Machine-learning systems and methods for audio enhancement

Abstract

Aspects of the disclosed technology include machine-learning systems and methods for enhancing speech signals. A machine-learned speech enhancement model can be trained using audio data and synthetic accelerometer data that is generated using a synthetic accelerometer data generation model. The accelerometer data can include synthetic noise data that can be used to train the speech enhancement model to learn a transfer function that models environmental noise to an accelerometer of a wearable device. By simulating data from multiple sensors, the disclosed technology provides a cost-effective solution for training robust large-scale models for speech enhancement.

Description

MACHINE-LEARNING SYSTEMS AND METHODS FOR AUDIO ENHANCEMENT FIELD

[0001] The present disclosure relates generally to machine learning processes and machine-learned devices and systems. More particularly, the present disclosure relates to machine-learning for audio enhancement.BACKGROUND

[0002] Artificial intelligence systems increasingly include large foundational machine-learned models which have the capability to provide a wide range of new product experiences. As an example, machine-learned generative models have proven successful at generating content including text, images, video, audio, computer-executable code, etc. Recently, machine-learned generative models have been used in speech enhancement systems to generate enhanced or denoised audio signals from input audio signals.

[0003] While machine-learned generative models have improved the capabilities of speech enhancement systems, traditional generative models may not adequately model all audio sources, leading to underperformance in model effectiveness.SUMMARY

[0004] Aspects and advantages of embodiments of the present disclosure will be set forth in part in the following description, or can be learned from the description, or can be learned through practice of the embodiments.

[0005] One example aspect of the present disclosure is directed to a computer-implemented method performed by a computing system including one or more computing devices. The method includes combining audio data representing at least one noise as detected by at least one microphone of a wearable device with audio data representing user speech as detected by the at least one microphone of the wearable device, combining synthetic accelerometer data representing the at least one noise as detected by at least one accelerometer of the wearable device with synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device, providing the combined audio data and the combined synthetic accelerometer data to a machine-learned speech enhancement model configured to generate an enhanced audio output based on audio data and accelerometer data, processing the combined audio data and the combined synthetic accelerometer data to generate an output including an enhanced audio speech signal,generating a loss based on comparing the enhanced audio speech signal with the audio data representing the user speech as detected by the at least one microphone of the wearable device, and modifying at least a portion of the machine-learned speech enhancement model based at least in part on the loss.

[0006] Another example aspect of the present disclosure is directed to a computing system including one or more processors, and one or more non-transitory computer-readable storage media that collectively store instructions that, when executed by the one or more processors, cause the one or more processors to perform operations. The operations include combining audio data representing at least one noise as detected by at least one microphone of a wearable device with audio data representing user speech as detected by the at least one microphone of the wearable device, combining synthetic accelerometer data representing the at least one noise as detected by at least one accelerometer of the wearable device with synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device, providing the combined audio data and the combined synthetic accelerometer data to a machine-learned speech enhancement model configured to generate an enhanced audio output based on audio data and accelerometer data, processing the combined audio data and the combined synthetic accelerometer data to generate an output including an enhanced audio speech signal, generating a loss based on comparing the enhanced audio speech signal with the audio data representing the user speech as detected by the at least one microphone of the wearable device, and modifying at least a portion of the machine-learned speech enhancement model based at least in part on the loss..

[0007] Yet another example aspect of the present disclosure is directed to one or more non-transitory computer-readable storage media that store instructions that, when executed by one or more processors, cause the one or more processors to perform operations. The operations include combining audio data representing at least one noise as detected by at least one microphone of a wearable device with audio data representing user speech as detected by the at least one microphone of the wearable device, combining synthetic accelerometer data representing the at least one noise as detected by at least one accelerometer of the wearable device with synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device, providing the combined audio data and the combined synthetic accelerometer data to a machine-learned speech enhancement model configured to generate an enhanced audio output based on audio data and accelerometer data, processing the combined audio data and the combined synthetic accelerometer data to generate an output including an enhanced audio speech signal, generating a loss based oncomparing the enhanced audio speech signal with the audio data representing the user speech as detected by the at least one microphone of the wearable device, and modifying at least a portion of the machine-learned speech enhancement model based at least in part on the loss.

[0008] Other example aspects of the present disclosure are directed to other systems, methods, apparatuses, tangible non-transitory computer-readable media, and devices for performing functions described herein. These and other features, aspects, and advantages of various implementations will become better understood with reference to the following description and appended claims. The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate implementations of the present disclosure and, together with the description, help explain the related principles.BRIEF DESCRIPTION OF THE DRAWINGS

[0009] FIG. 1 is a block diagram depicting an example computing environment including a machine learning system including a machine-learned speech enhancement model and a machine-learned downstream task model according to example embodiments of the present disclosure;

[0010] FIG. 2 is a block diagram depicting an example computing environment including a machine-learning system including a machine-learned speech enhancement model and a training engine for training the speech enhancement model using audio data and accelerometer data according to example embodiments of the present disclosure;

[0011] FIG. 3 is a block diagram depicting an example computing environment including a machine-learning system including an encoder / decoder architecture for a machine-learned speech enhancement model according to example embodiments of the present disclosure;

[0012] FIG. 4 is a block diagram depicting an example computing environment including a synthetic accelerometer data generation model according to example embodiments of the present disclosure;

[0013] FIG. 5 is a block diagram depicting an example computing environment including a machine-learning system including a machine-learned synthetic accelerometer data generation model and a training engine for training the synthetic accelerometer data generation model using audio data and accelerometer data according to example embodiments of the present disclosure;

[0014] FIG. 6 is a block diagram depicting a wearable device according to example embodiments of the present disclosure;

[0015] FIG. 7 is a flowchart diagram depicting an example method of training a machine-learned speech enhancement model according to example embodiments of the present disclosure;

[0016] FIG. 8 is a flow chart diagram illustrating an example method of training a machine-learned model according to example embodiments of the present disclosure;

[0017] FIG. 9 is a block diagram of an example processing flow for using machine-learned model(s) to process input(s) to generate output(s) according to example embodiments of the present disclosure;

[0018] FIG. 10 is a block diagram of an example sequence processing model according to example embodiments of the present disclosure;

[0019] FIG. 11 is a block diagram of an example technique for populating an example input sequence for processing by a sequence processing model according to example embodiments of the present disclosure;

[0020] FIG. 12 is a block diagram of an example model development platform according to example embodiments of the present disclosure;

[0021] FIG. 13 is a block diagram of an example training workflow for training a machine-learned model according to example embodiments of the present disclosure;

[0022] FIG. 14 is a block diagram of an inference system for operating one or more machine-learned model(s) to perform inference according to example embodiments of the present disclosure;

[0023] FIG. 15 is a block diagram of an example networked computing system according to example embodiments of the present disclosure;

[0024] FIG. 16 is a block diagram of an example computing device according to example embodiments of the present disclosure; and

[0025] FIG. 17 is a block diagram of an example computing device according to example embodiments of the present disclosure.DETAILED DESCRIPTION

[0026] Reference now will be made in detail to embodiments, one or more examples of which are illustrated in the drawings. Each example is provided by way of explanation of the embodiments, not limitation of the present disclosure. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made to the embodiments without departing from the scope or spirit of the present disclosure. For instance, features illustrated or described as part of one embodiment can be used with another embodiment toyield a still further embodiment. Thus, it is intended that aspects of the present disclosure cover such modifications and variations.Qverview

[0027] Generally, the present disclosure is directed to machine-learning systems and more particularly, to machine-learning systems for enhancing audio signals, such as speech signals generated by a microphone of a wearable device such as an earbud or other head mounted device (HMD). In accordance with example embodiments of the disclosed technology, a speech enhancement system for a wearable device leverages multimodal sensor data, including audio data and accelerometer data, to generate an enhanced speech signal such as a denoised speech signal using one or more machine-learned generative models. The machine-learned generative model(s) can be trained using synthetic multimodal data to provide a cost-effective and high-quality solution that addresses the high costs of acquiring real multimodal data from real users.

[0028] An accelerometer of a wearable device can generate accelerometer data that can be processed by a speech enhancement system along with an input audio waveform to generate a denoised speech signal. Traditionally, it has been assumed that the transfer function from the environment to the accelerometer was insignificant compared to the user’s speech. As such, generative models of traditional speech enhancement systems have not been trained to model environmental noise such as noises that may be detected by an accelerometer of a wearable device.

[0029] According to an example aspect of the present disclosure, however, a speech enhancement system can include one or more generative model(s) that are configured to model an environmental transfer function from the environment to the accelerometer. It has been discovered that the transfer function from the environment to an accelerometer can be significant, even in comparison to the transfer function from the user’s speech to the accelerometer. In accordance with example embodiments of the present disclosure, including the environmental transfer function can lead to significant improvements in generating speech enhancements such as denoised speech signals when compared with systems that rely solely on a user speech accelerometer transfer function.

[0030] In accordance with an example implementation of the disclosed technology, a machine-learned generative model of a speech enhancement system can be trained using synthetic accelerometer data that is generated by processing audio data with a synthetic accelerometer data generation model. Importantly, the generative model(s) can be trained tomodel different ty pes of noise detected by an accelerometer of a wearable device. Contrary to the assumptions of traditional solutions, the transfer function from the environment to the accelerometer can be significant, even when compared with the speech of the user wearing the device. In accordance with example embodiments, the generative model(s) can be trained to model an environmental accelerometer transfer function which leads to significant improvements when compared with solutions that rely solely on a user speech accelerometer transfer function.

[0031] In accordance with an example implementation, simulated multimodal sensor data can be used to train one or more machine-learned generative models of a speech enhancement system to enhance single-channel speech processing. The system can generate synthetic data from multiple simulated sensors to train large-scale generative models that effectively use the diverse information available from different channels. By utilizing simulated multimodal sensor data, the system can mitigate the cost and complexity associated with real-world data collection.

[0032] According to an example implementation, a machine-learned generative model of a speech enhancement system can be trained using a combination of audio data and synthetically-generated accelerometer data. The audio data can include audio data representing human speech in example embodiments. The audio data can include clean speech samples. The audio data can additionally include a diverse collection of environmental noises. For example, the environmental noises can include recordings of street noise, office noise, and natural ambient sounds that enable the generation of robust and varied speech samples.

[0033] A clean speech sample can be combined with one or more environmental noises to generate a noisy speech sample including the environmental noises. Additionally, a clean speech sample can be combined with another clean speech sample to generate a noisy speech sample including background or other speech in addition to the user speech. The environmental noises can be added at varying signal-to-noise ratios (e.g., ranging from approximately -20db to +50db) to simulate real-world noisy conditions that can provide diverse training scenarios for a model. Other ranges can be used such as from -lOdb to +50db, +20 to -50db, -20db to +40db, etc.

[0034] According to an example implementation, sensor simulation can be provided by simulating recordings from multiple microphones by creating spatially distributed noise and speech signals. A combined audio signal (e.g., speech + noise) can be processed to mimic the effects of different sensor placements and orientations. The characteristics of sensors,including typical amplitude response, can be measured and applied to enhance the realism of the simulated data.

[0035] Synthetic accelerometer data can be generated from the audio data and used for training the machine-learned generative model. In an example implementation, synthetic accelerometer data can be generated from the audio data using one or more synthetic accelerometer data generation models. In one example, the model can be a rule-based system that utilizes one or more rules and / or filters to generate synthetic accelerometer data. In another example, the data generation model can include a machine-learned generative model that is trained to generate accelerometer data from input audio data. In an example implementation, the generative model can include a transformer having one or more encoders and one or more decoders. In some examples, more than one generative model may be used. For example, a first generative model can be used to generate accelerometer data from speech audio data and a second generative model can be used to generate accelerometer data from environmental noise audio data.

[0036] Similar to the audio data, the accelerometer data can undergo noise addition to simulate real-world noisy conditions. For example, clean speech audio data samples can be provided to the model to generate clean speech accelerometer data samples. Similarly, accelerometer data for a collection of audio environmental noises can be generated. A clean speech accelerometer sample can be combined with one or more environmental noise accelerometer samples to generate a noisy speech accelerometer sample including the environmental noises. Similarly, a clean speech accelerometer sample can be combined with another clean speech accelerometer sample to generate a noisy speech sample including background or other speech in addition to the user speech.

[0037] As with the audio data, sensor simulation can be provided for the accelerometer data by simulating recordings from multiple accelerometers by creating spatially distributed noise and speech accelerometer signals. A combined accelerometer signal (e.g., speech + noise) can be processed to mimic the effects of different sensor placements and orientations. The characteristics of sensors, including typical amplitude and accelerometer response, can be measured and applied to enhance the realism of the simulated data.

[0038] According to an example aspect of the disclosed technology, a machine-learned speech enhancement model and a downstream machine-learned task model can be trained end-to-end. For example, a downstream task model can include a sequence processing model such as a large language model or other generative model. The system can be trained end-to-end by determining a loss from the downstream model’s predictions and backpropagating theloss through the network including the speech enhancement model and the downstream task model. In this manner, the frontend model can learn the specific enhancements for optimizing the performance of the downstream task model.

[0039] Systems and methods in accordance with example embodiments of the present disclosure provide a number of technical effects and benefits. A machine-learning system in accordance with example embodiments of the present disclosure can effectively enhance speech signals generated by microphones of wearable devices such as earbuds or head mounted devices (HMDs) by leveraging multimodal sensor data, including audio data and accelerometer data, as inputs into one or more machine- learned speech enhancement models. In accordance with the present disclosure, including an environmental transfer function in a speech enhancement model that operates on accelerometer data can lead to significant improvements in generating denoised speech signals compared to systems that rely solely on a speech accelerometer transfer function. By incorporating transfer functions from the environment to the accelerometer, the machine-learned speech enhancement model can be trained to model multiple types of noise that may be detected by the accelerometer of a w earable device.

[0040] Moreover, systems and methods in accordance with example embodiments of the present disclosure provide significant technical effects and benefits by reducing the computational cost of training large-scale speech enhancement models. The high cost of acquiring real multimodal data is mitigated by generating synthetic multimodal data from readily available open-source datasets. This synthetic data generation process, involving noise addition and sensor simulation, significantly reduces data acquisition costs.Furthermore, the use of simulated data allow s for the generation of a large and diverse dataset, leading to improved model robustness and generalization. The simplified approach using a model for the environmental accelerometer transfer function further reduces computational complexity and resource demands compared to prior art methods that rely on complex models for all aspects of the system. This can result in reduced training time and computational resources required for model development and deployment. Thus, a speech enhancement system in accordance with the present disclosure can mitigate the cost and complexity associated with real-world data collection while still providing training data for large-scale generative models that can effectively use the diverse information available from different channels. For example, an environmental accelerometer transfer function can lead to significant improvements when compared with systems that rely solely on a speech accelerometer transfer function.

[0041] Much of the following disclosure refers to large language models and other sequence processing models as specific examples of machine-learned generative models but it will be appreciated that the disclosure is equally applicable to any type of generative model such as other types of sequence processing models. For example, the disclosed technology can be used with large image models, multimodal models, and other types of foundational models. For instance, the generative models can operate in domains other than the audio domain, such as image domains, audio domains, biochemical domains, etc. For instance, a sequence processing model may be used to process sequential inputs for robotic controls and other tasks. Similarly, the generative model and / or the downstream applications can be configured to perform any number of tasks. For instance, if the inputs to the generative model and / or a downstream application are images or features that have been extracted from images, the output generated by the generative model for a given image can be scores for each of a set of object categories, with each score representing an estimated likelihood that the image contains an image of an object belonging to the category. As another example, if the inputs to the generative model and / or a downstream application are sensor data, the outputs can be robotic control signals.

[0042] With reference now to the Figures, example embodiments of the present disclosure will be discussed in further detail.Example System and Model Arrangements

[0043] FIG. 1 depicts a block diagram of an example computing environment 100 according to example embodiments of the present disclosure. The computing environment 100 includes a machine-learned speech enhancement model 102 and a machine-learned downstream task model 110. Speech enhancement model 102 is configured for processing audio data 106 and accelerometer 104 to generate an enhanced speech signal 108 such as a denoised speech signal. The enhanced speech signal can be provided to one or more downstream task models 110 to generate one or more task outputs.

[0044] The machine-learned speech enhancement model 102 and downstream task model 110 can be implemented by one or more computing systems. For example, the machine-learned speech enhancement model 102 and downstream task model 110 can be implemented by a user computing device such as a wearable device (e.g., earbud). In another example, the machine-learned speech enhancement model 102 and downstream task model 110 can be implemented by a computing device remote from a wearable device (e.g., earbud) such as a server computing system. In yet another example, the speech enhancement model 102 anddownstream task model 110 can be implemented by different computing systems. For instance, computing environment 100 may be implemented as a client server computing environment, including one or more client computing devices and one or more server computing devices.

[0045] It will be appreciated that the term “system” can refer to specialized hardware, computer logic that executes on a more general processor, or some combination thereof. Thus, a system can be implemented in hardware, application specific circuits, firmware, and / or software controlling a general-purpose processor. In one embodiment, the systems can be implemented as program code files stored on a storage device, loaded into memory and executed by a processor or can be provided from computer program products, for example computer executable instructions, that are stored in a tangible computer-readable storage medium such as RAM, hard disk, or optical or magnetic media.

[0046] Speech enhancement model 102 can be implemented as one or more machine-learned generative models in example embodiments. Speech enhancement model 102 can include various types of machine-learned models. In an example embodiment, speech enhancement model 102 can include a machine-learned generative model such as a transformer that employs an encoder / decoder architecture. A generative model can include any ty pe of machine-learned generative model. In an example, a generative model can include a sequence processing model, such as a large language model including 10B parameters or more. In another example, a generative model can include a language model having less than 10B parameters (e g., IB parameters). In yet another example, the generative model can include an autoregressive language model or an image diffusion model.

[0047] Similar to speech enhancement model 102, downstream task model 110 can be implemented as one or more machine-learned generative models. In an example embodiment, downstream task model 110 can include a sequence processing model such as a large language model or other generative model. Example downstream task models 110 could also include, but are not limited to, Convolutional Neural Networks (CNN), Long short-term Memory’ Networks (LSTM), Recurrent Neural Networks (RNN), Bidirectional RNN Networks (BRNN), Generative Adversarial Networks (GAN), and Transformers. A generative downstream task model can include a machine-learned text-to-image model, a machine-learned text-to-video model, a machine-learned text-to-audio model, a machine-learned multi-modal model, or any other machine-learned model configured to provide generative content in response to an input. The generative content generated by generative models can include computer-executable code data, text data, image data, video data, audiodata, or other types of generative content. The generative model can be trained to process input data to generate output data.

[0048] Speech enhancement model 102 is configured to receive one or more inputs including audio data 106 such as microphone data generated by one or more audio sensors such as a microphone of a wearable device. Speech enhancement model 102 is also configured to receive one or more inputs including accelerometer data 104 such as accelerometer data generated by one or more accelerometers of the wearable device. A voice accelerometer (VA) can be incorporated into a wearable device such as earbuds. The voice accelerometer can detect a user speaking based on sound that travels by means of bone conduction. The voice accelerometer may detect a user's voice through bone conduction exhibited and detectable as vibration of the bony structure of the user.

[0049] Speech enhancement model 102 is configured to process accelerometer data (e.g., accelerometer data 104) and audio data (e.g., audio data 106) in order to output an enhanced speech signal 108 (e.g., enhanced speech data). Enhanced speech data 108 can include enhanced audio data representing denoised speech audio data. The accelerometer data 104 acts as a conditioning signal for the audio data 106. While speech processing has advanced with the development of sophisticated models and algorithms designed to enhance and recognize single channel speech signals, there can be instances when a single channel at a low signal to noise ratios (SNR) may not perform well, regardless of the amount of data. In both mask-based enhancement and generative enhancements, and regardless of the optimized performance of the model, the restored or estimated signal may not enclose the full information of the clean source signal in low SNR regions. Multimodal data, such as that obtained from one or more microphones and one or more accelerometers, can provide rich information that significantly enhances the performance of speech processing systems.

[0050] The enhanced voice audio data can be provided to one or more downstream task models such as a machine-learned sequence processing model (e.g., LLM). The downstream task model can vary7and include any suitable downstream task model for performing any number of tasks.

[0051] FIG. 2 is a block diagram depicting an example computing environment 200 including a machine-learning system in accordance with example embodiments of the present disclosure. Computing environment 200 includes a machine-learned speech enhancement model 202 and a training engine 220 for training the speech enhancement model 202 using audio data and accelerometer data according to example embodiments of the present disclosure. Training engine 220 can include a training computing system implementedlocally at a wearable device and / or remotely at one or more remote computing systems such as a mobile computing device (e.g., smartphone, etc.) or server computing system.

[0052] Speech enhancement model 202 can be trained using training data that includes a noisy speech audio sample and noisy accelerometer sample as inputs and a clean speech sample as a target output. Speech enhancement model 202 can receive multimodal inputs including audio signals and accelerometer signals. During training, speech enhancement model receives one or more inputs including audio data (e.g.. microphone data) and accelerometer data. An audio data input sample can be generated by combining noise audio data 212 with voice audio data 214. The voice audio data 214 can include an audio recording or sample of clean speech in an example implementation. The noise audio data 212 can include an audio recording or sample including noise (e.g., environmental noise or another speaker). The noise audio data and the voice audio data can be combined using various mixing and / or combination technologies. The combined noise audio data sample and voice audio data sample can be provided as an input to the speech enhancement model 202.

[0053] Speech enhancement model 202 can also receive an input including synthetic accelerometer data . The synthetic accelerometer data can be generated from audio data using a synthetic accelerometer data generation model. The synthetic accelerometer data input can be generated by combining noise accelerometer data 216 with voice accelerometer data 218. The noise accelerometer data can include an accelerometer recording or sample of noise (e.g.. environmental noise or another speaker). The voice accelerometer data can include an accelerometer recording or sample of clean speech in an example implementation. The noise accelerometer data and the voice accelerometer data can be combined using various mixing and / or combination technologies. The combined noise accelerometer data sample can be provided as an input to the speech enhancement model 202.

[0054] Speech enhancement model 202 is configured to receive the inputs including audio data (e.g., audio data 212 and 214) and accelerometer data (e.g., accelerometer data 216 and 218) and generate an output including enhanced speech signal 208. Enhanced speech signal 208 can include enhanced or denoised speech data such as audio data representing denoised speech audio data.

[0055] The enhanced speech signal 208 is provided to training engine 220. Additionally, the voice audio data 214 (without noise) is provided to the training engine 220. Training engine can compare the enhanced speech data 208 to the clean speech sample 214 to determine a loss 222. Training engine 220 can backpropagate the loss 222 through the speech enhancement model 202. The model 202 can be updated based on thebackpropagated loss 222. Example losses 222 can include mean square error (MSE) losses, cross-entropy losses, adversarial losses, discriminator losses, etc.

[0056] As depicted in FIG. 2, the prohibitive cost and logistical challenges of acquiring real multimodal data which often limit the training of large-scale generative models can be overcome by training a speech enhancement model using synthetic accelerometer data. Moreover, the accelerometer data can include data from simulated sensors which can be generated without requiring the acquisition of real data from real users. Furthermore, by incorporating an environmental accelerometer transfer function into the frontend enhancement model and utilizing synthetic microphone data from simulated sensors, a cost-effective and practical solution is provided for training a robust speech enhancement model 202. Speech enhancement model 202 can be trained on this synthetic multimodal data and can effectively utilize the additional information provided by multiple channels, leading to improved performance by the speech enhancement model as well as downstream task models. In example implementations, this multichannel approach not only enhances speech command recognition but also maintains key performance indicator (KPI) thresholds at significantly lower signal to noise ratios (SNRs) compared to single-channel methods. This underscores the efficacy of integrating diverse data sources for speech enhancement tasks.

[0057] In example embodiments, the integration of a downstream task model w ith the frontend speech enhancement model 202 through end-to-end training provides specific technical effects and benefits. By backpropagating the loss from the downstream task through the entire system, the frontend enhancement model can learn the precise enhancements needed to optimize the performance of the downstream task. This end-to-end training paradigm provides that the enhancements made to the speech signal are tailored to the specific requirements of the downstream task, which can lead to superior overall performance.

[0058] FIG. 3 is a block diagram depicting an example computing environment 300 including a machine-learning system including an encoder / decoder architecture for a machine-learned speech enhancement model according to example embodiments of the present disclosure. As described in FIG. 2, an audio input including a combination of noise audio data 312 and voice audio data 314 and an accelerometer input including a combination of noise accelerometer data 316 and voice accelerometer data 318 can be provided to a speech enhancement model 302 including an encoder 332 and decoder 334. The speech enhancement model 302 can include a symmetric encoder-decoder network with skip connections. The encoder and decoder can each include multiple blocks between convolutionlayers. The encoder can follow a downsampling scheme and the decoder can up-sample in the reverse order. The encoder encodes the multimodal inputs and the decoder generates an output including an enhanced speech signal. The enhanced speech signal can include enhanced speech data 308 such as denoised voice data.

[0059] The voice audio data 314 and the enhanced speech data 308 are provided to discriminator 336. Discriminator 336 can include a multi-resolution convolutional architecture. One or more discriminators can be used for input audio at different resolutions. The discriminator 336 can include multiple plain convolution layers followed by multiple grouped convolutions which are followed by multiple plain convolutional layers to produce an output. The speech enhancement model 302 and the discriminator 336 can be trained simultaneously using adversarial and reconstruction losses in an example implementation.

[0060] FIG. 4 is a block diagram depicting an example computing environment including a synthetic accelerometer data generation model 452 according to example embodiments of the present disclosure. Synthetic accelerometer data generation model 452 is configured to receive an input including audio data 406 and generate an output including synthetic accelerometer data 404. Synthetic accelerometer data generation model 452 can be a trained model that can receive an input audio data sample and generate synthetic accelerometer data based on the audio data sample. In another example, synthetic accelerometer data generation model 452 can be a non-machine-leamed model including one or more rules or heuristics, filters, or other logic or algorithms that can generate synthetic accelerometer data based on an audio data input. Synthetic accelerometer data generation model can include a symmetric encoder-decoder network in example embodiments.

[0061] In accordance with example embodiments of the present disclosure, multiple synthetic accelerometer data generation models can be used to generate synthetic accelerometer data from audio sources. For example, synthetic accelerometer data generation model 452 can include a first generative model to generate synthetic accelerometer data from speech audio data and a second generative model to generate synthetic accelerometer data from environmental noise audio data. In some examples, different generative models can be used to generate synthetic accelerometer data from different noise sources. For example, a particular generative model can be used to generate synthetic accelerometer data from audio of street noises, office noises, and natural ambient sounds, etc. while a different generative model can be used to generate synthetic accelerometer data for wind noises.

[0062] FIG. 5 is a block diagram depicting an example computing environment including a machine-learning system including a machine-learned synthetic accelerometer datageneration model 552 and a training engine 520 for training the synthetic accelerometer data generation model 552 using audio data and accelerometer data according to example embodiments of the present disclosure. The training data can include an input that includes audio data 506 (e.g., speech audio, noise audio, etc.) and a target that includes accelerometer data 504 corresponding to the audio data 506. The audio data 506 can be provided as input to the speech enhancement model 502. In response to the audio data 506, the speech enhancement model 502 can generate an output including synthetic accelerometer data 554. The synthetic accelerometer data 554 can be provided to the training engine 520.Additionally, the target accelerometer data 554 can be provided to the training engine 520. Training engine 520 can compare the synthetic accelerometer data 554 to the accelerometer data 504 to determine a loss 522. Training engine 520 can backpropagate the loss 522 through the synthetic accelerometer data generation model 552. The model 552 can be updated based on the backpropagated loss 522. Example losses 522 can include mean square error (MSE) losses, cross-entropy losses, etc.

[0063] FIG. 6 is a block diagram depicting an example wearable device 602 in accordance with example embodiments of the present disclosure. Examples of wearable devices can include earbuds, other head mounted devices, watches, rings, and other devices wearable by a user. Wearable device 602 includes processing circuitry7604 (e.g., processor, microprocessor, etc.), memory 606 (e.g., RAM and / or ROM) storing data 608 and instructions 612. input device(s) 614, output device(s) 616, microphone(s) 618, accelerometer(s) 620, network interface(s) 622 (e.g., Bluetooth, WiFi, USB, etc ), and power source 624. Wearable device 202 is one example of a wearable device as described herein. It will be appreciated while specific components are depicted in FIG. 2, additional or fewer components may be included in a wearable device in accordance with example embodiments of the present disclosure.

[0064] Processing circuitry 604 can include one or more electric circuits that comprise one or more processors such as one or more microprocessors. Memory 606 can include (e.g., store, and / or the like) data 608 and instructions 612. When executed by processing circuitry 604, instructions stored in memory 606 can cause processing circuitry 604 to perform one or more operations, functions, and / or the like described herein. Data 608 can include one or more machine-learned models such as a machine-learned speech enhancement model in example embodiments.

[0065] Wearable device 602 may include one or more input devices 614 and one or more output devices 616. An input device such as a touch input device can be utilized to enableuser to provide input to the wearable device. An output device such as a haptic device, display, or speaker can be utilized to enable user to receive the output from the wearable device.

[0066] An output device can be configured to provide a haptic response, a tactical response, an audio response, a visual response, or some combination thereof. Output devices may include visual output devices, such as one or more light-emitting diodes (LEDs), audio output devices such as one or more speakers, one or more tactile output devices, and / or one or more haptic output devices. In some examples, the one or more output devices are formed as part of the wearable device, although this is not required. In one example, an output device can include one or more devices configured to provide different types of haptic output signals. For example, the one or more haptic devices can be configured to generate specific output signals in the form of different vibrations and / or vibration patterns.

[0067] Wearable device 602 can include one or more microphones 618 or other audio sensors configured to generate audio signals including audio data. Wearable device 602 can include one or more accelerometer(s) configured to generate accelerometer signals including accelerometer data. The accelerometer 620 can generate sensor data including accelerometer data or inertial data indicative of a position, velocity, and / or an acceleration of the interactive object. In example embodiments, accelerometer 620 can include or be integrated with an inertial measurement unit. Accelerometer 620 can include a voice accelerometer that functions as a vibration sensor to provide an inertial signal based on bone conductance. For example, accelerometer 620 can be configured to measure vibrations of a user’s head to represent an utterance by the user. The accelerometer 620 may generate one or more outputs describing one or more motions (e.g., 2-D or 3-D) of the wearable device 602. The accelerometer 620 may be secured to the device, for example, with zero degrees of freedom, either removably or irremovably, such that the accelerometer translates and is reoriented as the wearable device 602 is translated and are reoriented. In some embodiments, the accelerometer can be integrated with an inertial measurement unit that may include a gyroscope or an accelerometer (e.g., a combination of a gyroscope and an accelerometer), such as a three axis gyroscope or accelerometer configured to sense rotation and acceleration along and about three, generally orthogonal axes. In some embodiments, the inertial measurement unit(s) may include a sensor configured to detect changes in velocity or changes in rotational velocity of the interactive object and an integrator configured to integrate signals from the sensor such that a net movement may be calculated, for instance bya processor of the inertial measurement unit, based on an integrated movement about or along each of a plurality of axes.

[0068] Network interface 622 can enable wearable device 602 to communicate with one or more computing devices. By way of example and not limitation, network interfaces 622 may communicate data over a local-area-network (LAN), a wireless local-area-network (WLAN), a personal-area-network (PAN) (e.g., Bluetooth™), a wide-area-network (WAN), an intranet, the Internet, a peer-to-peer network, point-to-point network, a mesh network, and the like. Network interface 622can be a wired and / or wireless network interface. Power source 624 may be coupled, via one or more interfaces to provide power to the various components of the wearable device, and may be implemented as a small batten' in some examples.

[0069] Wearable device 602 can include various other types of electronics, such as additional sensors (e.g., capacitive touch sensors, microphones, accelerometers, ambient temperature sensor, barometer, ECG, EDA, PPG), output devices (e.g., LEDs, speakers, or haptic devices), electrical circuitry, and so forth. The various electronics depicted within wearable device 602 may be physically and permanently embedded within wearable device 602 in example embodiments. In some examples, one or more components may be removably coupled to the wearable device 602. By way of example, a removable power source 624 may be included in example embodiments. In some examples, the internal electronics of the wearable device 602 can include a flexible printed circuit board (PCB).

[0073] While wearable device 602 is illustrated and described as including specific electronic components, it will be appreciated that wearable devices may be configured in a variety of different ways. For example, in some cases, electronic components described as being contained within a wearable device may at least be partially implemented at another computing device, and vice versa. Furthermore, wearable device 602 may include electronic components other that those illustrated in FIG. 6, such as sensors, light sources (e.g., LED’s), displays, speakers, and so forth.Example Methods

[0070] FIG. 7 is a flowchart diagram depicting an example method 600 of training a machine-learned speech enhancement model using synthetic accelerometer data according to example embodiments of the present disclosure. One or more portion(s) of example method 700 and the other methods described here can be implemented by a computing system that includes one or more computing devices such as, for example, a machine-learned computingsystem as described herein. Each respective portion of example method 700 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 700 can be implemented on the hardware components of the device(s) described herein, for example, to train a machine-learned speech enhancement model. FIG. 7 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. FIG. 7 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of example method 700 can be performed additionally, or alternatively, by other systems.

[0071] At 702, example method 700 can include combining audio data representing at least one noise as detected by at least one microphone of a wearable device with audio data representing user speech as detected by the at least one microphone. The audio data representing user speech can include audio data samples representing audio detected by one or more audio sensors of a wearable device. Audio data can include speech audio data in an example implementation, as well as other forms of audio data. The audio data representing noise can include audio data samples representing audio of one or more environmental noises as detected by the audio sensor(s). For example, environmental noises can include audio data samples representing street noise, office noise, construction noise, nature noises, and other environmental noises. In some examples, a user speech sample can be combined with another user speech sample to represent background speaker noise.

[0072] At 704, example method 700 can include combining synthetic accelerometer data representing the at least one noise as detected by at least one accelerometer of the wearable device with synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device. In some examples, a synthetic accelerometer data generation model can be used to generate synthetic accelerometer data including synthetic accelerometer data representing noise and synthetic accelerometer data representing user speech. The synthetic accelerometer data can be generated by inputting corresponding audio data samples into a synthetic accelerometer data generation model. The accelerometer data representing noise can be combined with the speech accelerometer data for training the model to leam to enhance speech with noise in it. The accelerometer data representing noise(or system noise) can be used to leam a transfer function from the environment to the accelerometer.

[0073] At 706, example method 700 can include providing the combined audio data and the combined synthetic accelerometer data to a machine-learned speech enhancement model configured to generate an enhanced audio output based on audio data and accelerometer data. The combined audio data representing noise may be input into the machine-learned speech enhancement model (e.g., a machine-learned generative model). The combined synthetic accelerometer data representing noise may also be input into the machine-learned speech enhancement model 708.

[0074] At 708, example method 700 can include processing the combined audio data and the combined synthetic accelerometer data to generate an output including an enhanced audio speech signal. The output can include an enhanced audio speech signal that corresponds to the user’s speech (e.g., an enhanced or denoised voice signal).

[0075] At 710, example method 700 can include generating at least one loss based on comparing the enhanced audio speech signal with the audio data representing the user speech. The loss can include mean square error (LSE), cross-entropy losses, adversarial losses, discriminator losses, etc.

[0076] At 712, example method 700 can include modifying at least a portion of the machine-learned speech enhancement model based at least in part on the loss. Modifying the machine-learned speech enhancement model can include modifying the machine-learned speech enhancement model to improve generation of the enhanced audio speech signal using the combined audio data and the combined synthetic accelerometer data. Modify ing the machine-learned speech enhancement model can include modifying one or more weights or parameters of the machine-learned speech enhancement model. Modifying the machine-learned speech enhancement model can include modifying at least a portion of the machine-learned speech enhancement model including at least an encoder or a decoder, in an example implementation.

[0077] FIG. 8 depicts a flowchart of a method 800 for training one or more machine-learned models according to aspects of the present disclosure. For instance, an example machine-learned model can include a core sequence processing model, such as a foundational large language model (LLM).

[0078] At 802, example method 800 can include obtaining a training instance. A set of training data can include a plurality of training instances divided between multiple datasets (e.g., a training dataset, a validation dataset, or testing dataset). A training instance can belabeled or unlabeled. Although referred to in example method 800 as a “training"’ instance, it is to be understood that runtime inferences can form training instances when a model is trained using an evaluation of the model’s performance on that runtime instance (e.g., online training / leaming). Example data types for the training instance and various tasks associated therewith are described throughout the present disclosure.

[0079] At 804, example method 800 can include processing, using one or more machine-learned models, the training instance to generate an output. The output can be directly obtained from the one or more machine-learned models or can be a downstream result of a chain of processing operations that includes an output of the one or more machine-learned models.

[0080] At 806. example method 800600 can include receiving an evaluation signal associated with the output. The evaluation signal can be obtained using a loss function. Various determinations of loss can be used, such as mean squared error, likelihood loss, cross entropy loss, hinge loss, contrastive loss, or various other loss functions. The evaluation signal can be computed using known ground-truth labels (e g., supervised learning), predicted or estimated labels (e.g.. semi- or self-supervised learning), or without labels (e.g.. unsupervised learning). The evaluation signal can be a reward (e.g., for reinforcement learning). The reward can be computed using a machine-learned reward model configured to generate rewards based on output(s) received. The reward can be computed using feedback data describing human feedback on the output(s).

[0081] At 808, example method 800 can include updating the machine-learned model using the evaluation signal. For example, values for parameters of the machine-learned model(s) can be learned, in some embodiments, using various training or learning techniques, such as, for example, backwards propagation. For example, the evaluation signal can be backpropagated from the output (or another source of the evaluation signal) through the machine-learned model(s) to update one or more parameters of the model(s) (e.g., based on a gradient of the evaluation signal with respect to the parameter value(s)). For example, system(s) containing one or more machine-learned models can be trained in an end-to-end manner. Gradient descent techniques can be used to iteratively update the parameters over a number of training iterations. In some implementations, performing backwards propagation of errors can include performing truncated backpropagation through time. Example method 600 can include implementing a number of generalization techniques (e.g., weight decays, dropouts, etc.) to improve the generalization capability of the models being trained.

[0082] In some implementations, example method 800 can be implemented for training a machine-learned model from an initialized state to a fully trained state (e.g., when the model exhibits a desired performance profile, such as based on accuracy, precision, recall, etc.).

[0083] In some implementations, example method 800 can be implemented for particular stages of a training procedure. For instance, in some implementations, example method 800 can be implemented for pre-training a machine-learned model. Pre-training can include, for instance, large-scale training over potentially noisy data to achieve a broad base of performance levels across a variety of tasks / data types. In some implementations, example method 600 can be implemented for fine-tuning a machine-learned model. Fine-tuning can include, for instance, smaller-scale training on higher-quality (e.g., labeled, curated, etc.) data. Fine-tuning can affect all or a portion of the parameters of a machine-learned model. For example, various portions of the machine-learned model can be “frozen” for certain training stages. For example, parameters associated with an embedding space can be “frozen” during fine-tuning (e.g., to retain information learned from a broader domain(s) than present in the fine-tuning dataset(s)). An example fine-tuning approach includes reinforcement learning. Reinforcement learning can be based on user feedback on model performance during use.Example Machine-Learned Models

[0084] FIG. 9 is a block diagram of an example processing flow for using machine-learned model(s) 1 to process input(s) 2 to generate output(s) 3.

[0085] Machine-learned model(s) 1 can be or include one or multiple machine-learned models or model components. Example machine-learned models can include neural networks (e.g.. deep neural networks). Example machine-learned models can include non-linear models or linear models. Example machine-learned models can use other architectures in lieu of or in addition to neural networks. Example machine-learned models can include decision tree based models, support vector machines, hidden Markov models, Bayesian networks, linear regression models, k-means clustering models, etc.

[0086] Example neural networks can include feed-forward neural networks, recurrent neural networks (RNNs), including long short-term memory (LSTM) based recurrent neural networks, convolutional neural networks (CNNs), diffusion models, generative-adversarial networks, or other forms of neural networks. Example neural networks can be deep neural networks. Some example machine-learned models can leverage an attention mechanism, suchas self-atention. For example, some example machine-learned models can include multiheaded self-attention models.

[0087] Machine-learned model(s) 1 can include a single or multiple instances of the same model configured to operate on data from input(s) 2. Machine-learned model(s) 1 can include an ensemble of different models that can cooperatively interact to process data from input(s) 2. For example, machine-learned model(s) 1 can employ a mixture-of-experts structure. See. e.g., Zhou et al., Mixture-of-Experts with Expert Choice Routing, arXiv:2202.09368v2 (Oct.14, 2022).

[0088] Input(s) 2 can generally include or otherwise represent various types of data. Input(s) 2 can include one type or many different types of data. Output(s) 3 can be data of the same type(s) or of different types of data as compared to input(s) 2. Output(s) 3 can include one type or many different types of data.

[0089] Example data types for input(s) 2 or output(s) 3 include natural language text data, software code data (e.g., source code, object code, machine code, or any other form of computer-readable instructions or programming languages), machine code data (e.g.. binary code, assembly code, or other forms of machine-readable instructions that can be executed directly by a computer's central processing unit), assembly code data (e.g., low-level programming languages that use symbolic representations of machine code instructions to program a processing unit), genetic data or other chemical or biochemical data, image data, audio data, audiovisual data, haptic data, biometric data, medical data, financial data, statistical data, geographical data, astronomical data, historical data, sensor data generally (e.g., digital or analog values, such as voltage or other absolute or relative level measurement values from a real or artificial input, such as from an audio sensor, light sensor, displacement sensor, etc.), and the like. Data can be raw or processed and can be in any format or schema.

[0090] In multimodal inputs 2 or outputs 3, example combinations of data types include image data and audio data, image data and natural language data, natural language data and software code data, image data and biometric data, sensor data and medical data, etc. It is to be understood that any combination of data types in an input 2 or an output 3 can be present.

[0091] An example input 2 can include one or multiple data types, such as the example data types noted above. An example output 3 can include one or multiple data types, such as the example data types noted above. The data type(s) of input 2 can be the same as or different from the data type(s) of output 3. It is to be understood that the example data types noted above are provided for illustrative purposes only. Data types contemplated within the scope of the present disclosure are not limited to those examples noted above.Example Machine-Learned Sequence Processing Models

[0092] FIG. 10 is a block diagram of an example implementation of an example machine-learned model configured to process sequences of information. For instance, an example implementation of machine-learned model(s) 1 can include machine-learned sequence processing model(s) 4. An example system can pass input(s) 2 to sequence processing model(s) 4. Sequence processing model(s) 4 can include one or more machine-learned components. Sequence processing model(s) 4 can process the data from input(s) 2 to obtain an input sequence 5. Input sequence 5 can include one or more input elements 5-1, 5-2, . . . , 5-Af, etc. obtained from input(s) 2. Sequence processing model 4 can process input sequence 5 using prediction layer(s) 6 to generate an output sequence 7. Output sequence 7 can include one or more output elements 7-1, 7-2, . . . , 7-N. etc. generated based on input sequence 5. The system can generate output(s) 3 based on output sequence 7.

[0093] Sequence processing model(s) 4 can include one or multiple machine-learned model components configured to ingest, generate, or otherwise reason over sequences of information. For example, some example sequence processing models in the text domain are referred to as “Large Language Models / ’ or LLMs. See. e.g., PaLM 2 Technical Report. GOOG E, https: / / ai.google / static / documents / palm2techreport.pdf (n.d.). Other example sequence processing models can operate in other domains, such as image domains, see, e.g. , Dosovitskiy et al., An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, ARXIV:2010.11929V2 (Jun. 3. 2021), audio domains, see. e.g. , Agostinelli et al..MusicLM: Generating Music From Text. ARXIV:2301.11325vl (Jan. 26, 2023), biochemical domains, see, e.g. , Jumper et al., Highly accurate protein structure prediction with AlphaFold, 596 Nature 583 (Aug. 26, 2021), by way of example. Sequence processing model(s) 4 can process one or multiple types of data simultaneously. Sequence processing model(s) 4 can include relatively large models (e.g., more parameters, computationally expensive, etc.), relatively small models (e.g., fewer parameters, computationally lightweight, etc.), or both.

[0094] In general, sequence processing model(s) 4 can obtain input sequence 5 using data from input(s) 2. For instance, input sequence 5 can include a representation of data from input(s) 2 in a format understood by sequence processing model(s) 4. One or more machine-learned components of sequence processing model(s) 4 can ingest the data from input(s) 2, parse the data into pieces compatible with the processing architectures of sequence processing model(s) 4 (e.g., via “tokenization’’). and project the pieces into an input space associated with prediction layer(s) 6 (e.g.. via “embedding”).

[0095] Sequence processing model(s) 4 can ingest the data from input(s) 2 and parse the data into a sequence of elements to obtain input sequence 5. For example, a portion of input data from input(s) 2 can be broken down into pieces that collectively represent the content of the portion of the input data. The pieces can provide the elements of the sequence.

[0096] Elements 5-1, 5-2, . . . , 5-M can represent, in some cases, building blocks for capturing or expressing meaningful information in a particular data domain. For instance, the elements can describe ‘"atomic units” across one or more domains. For example, for textual input source(s), the elements can correspond to groups of one or more words or sub-word components, such as sets of one or more characters.

[0097] For example, elements 5-1. 5-2, . . . . 5-M can represent tokens obtained using a tokenizer. For instance, a tokenizer can process a given portion of an input source and output a series of tokens (e.g., corresponding to input elements 5-1, 5-2, . . . , 5-M) that represent the portion of the input source. Various approaches to tokenization can be used. For instance, textual input source(s) can be tokenized using a byte-pair encoding (BPE) technique. See, e.g., Kudo et al., SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing. PROCEEDINGS OF THE 2018 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (System Demonstrations), pages 66-71 (October 31-November 4, 2018), https: / / aclanthology.org / D18-2012.pdf. Image-based input source(s) can be tokenized by extracting and serializing patches from an image.

[0098] In general, arbitrary data types can be serialized and processed into input sequence 5. It is to be understood that element(s) 5-1, 5-2, . . . , 5-AL depicted in FIG. 7 can be the tokens or can be the embedded representations thereof.

[0099] Prediction layer(s) 6 can predict one or more output elements 7-1, 7-2, . . . , 7-N based on the input elements. Prediction layer(s) 6 can include one or more machine-learned model architectures, such as one or more layers of learned parameters that manipulate and transform the input(s) to extract higher-order meaning from, and relationships between, input element(s) 5-1, 5-2, . . . , 5-M. In this manner, for instance, example prediction layer(s) 6 can predict new output element(s) in view of the context provided by input sequence 5.

[0100] Prediction layer(s) 6 can evaluate associations between portions of input sequence 5 and a particular output element. These associations can inform a prediction of the likelihood that a particular output follows the input context. For example, consider the textual snippet, “The carpenter’s toolbox was small and heavy. It was full of .” Example prediction layer(s) 6 can identify that “It” refers back to “toolbox” by determining a relationship between the respective embeddings. Example prediction layer(s) 6 can also link“It” to the attributes of the toolbox, such as “small” and “heavy .” Based on these associations, prediction layer(s) 6 can, for instance, assign a higher probability to the word “nails” than to the word “sawdust.”

[0101] A transformer is an example architecture that can be used in prediction layer(s) 6. See, e.g. , Vaswani et al., Attention Is All You Need, ARXIV:1706.03762V7 (Aug. 2, 2023). A transformer is an example of a machine-learned model architecture that uses an attention mechanism to compute associations between items within a context window. The context win ow can include a sequence that contains input sequence 5 and potentially one or more output element(s) 7-1, 7-2, . . . , 7-N. A transformer block can include one or more attention layer(s) and one or more post-attention layer(s) (e g., feedforward layer(s), such as a multilayer perceptron).

[0102] Prediction layer(s) 6 can include other machine-learned model architectures in addition to or in lieu of transformer-based architectures. For example, recurrent neural networks (RNNs) and long short-term memoiv (LSTM) models can also be used, as well as convolutional neural networks (CNNs). In general, prediction layer(s) 6 can leverage various kinds of artificial neural networks that can understand or generate sequences of information.

[0103] Output sequence 7 can include or otherwise represent the same or different data types as input sequence 5. For instance, input sequence 5 can represent textual data, and output sequence 7 can represent textual data. Input sequence 5 can represent image, audio, or audiovisual data, and output sequence 7 can represent textual data (e.g., describing the image, audio, or audiovisual data). It is to be understood that prediction layer(s) 6, and any other interstitial model components of sequence processing model(s) 4, can be configured to receive a variety of data types in input sequence(s) 5 and output a variety of data types in output sequence(s) 7.

[0104] Output sequence 7 can have various relationships to input sequence 5. Output sequence 7 can be a continuation of input sequence 5. Output sequence 7 can be complementary to input sequence 5. Output sequence 7 can translate, transform, augment, or otherwise modify input sequence 5. Output sequence 7 can answer, evaluate, confirm, or otherwise respond to input sequence 5. Output sequence 7 can implement (or describe instructions for implementing) an instruction provided via input sequence 5.

[0105] Output sequence 7 can be generated autoregressively. For instance, for some applications, an output of one or more prediction layer(s) 6 can be passed through one or more output layers (e.g., softmax layer) to obtain a probability distribution over an output vocabulary (e.g., a textual or symbolic vocabulary) conditioned on a set of input elements in acontext window. In this manner, for instance, output sequence 7 can be autoregressively generated by sampling a likely next output element, adding that element to the context window, and re-generating the probability distribution based on the updated context window; and sampling a likely next output element, and so forth.

[0106] Output sequence 7 can also be generated non-autoregressively. For instance, multiple output elements of output sequence 7 can be predicted together without explicit sequential conditioning on each other. See, e.g.. Saharia et al., Non-Autoregressive Machine Translation with Latent Alignments, ARXIV:2004.07437V3 (NOV. 16, 2020).

[0107] Output sequence 7 can include one or multiple portions or elements. In an example content generation configuration, output sequence 7 can include multiple elements corresponding to multiple portions of a generated output sequence (e.g., a textual sentence, values of a discretized waveform, computer code, etc.). In an example classification configuration, output sequence 7 can include a single element associated with a classification output. For instance, an output “vocabulary” can include a set of classes into which an input sequence is to be classified. For instance, a vision transformer block can pass latent state information to a multilayer perceptron that outputs a likely class value associated with an input image.

[0108] FIG. 11 is a block diagram of an example technique for populating an example input sequence 8. Input sequence 8 can include various functional elements that form part of the model infrastructure, such as an element 8-0 obtained from a task indicator 9 that signals to any model(s) that process input sequence 8 that a particular task is being performed (e.g., to help adapt a performance of the model(s) to that particular task). Input sequence 8 can include various data elements from different data modalities. For instance, an input modality 10-1 can include one modality of data. A data-to-sequence model 11-1 can process data from input modality 10-1 to project the data into a format compatible with input sequence 8 (e.g., one or more vectors dimensioned according to the dimensions of input sequence 8) to obtain elements 8-1, 8-2, 8-3. Another input modality 10-2 can include a different modality of data. A data-to-sequence model 11-2 can project data from input modality 10-2 into a format compatible with input sequence 8 to obtain elements 8-4, 8-5. 8-6. Another input modality 10-3 can include yet another different modality of data. A data-to-sequence model 11-3 can project data from input modality 10-3 into a format compatible with input sequence 8 to obtain elements 8-7, 8-8. 8-9.

[0109] Input sequence 8 can be the same as or different from input sequence 5. Input sequence 8 can be a multimodal input sequence that contains elements that represent datafrom different modalities using a common dimensional representation. For instance, an embedding space can have P dimensions. Input sequence 8 can be configured to contain a plurality of elements that have P dimensions. In this manner, for instance, example implementations can facilitate information extraction and reasoning across diverse data modalities by projecting data into elements in the same embedding space for comparison, combination, or other computations therebetween.

[0110] For example, elements 8-0. . . . . 8-9 can indicate particular locations within a multidimensional embedding space. Some elements can map to a set of discrete locations in the embedding space. For instance, elements that correspond to discrete members of a predetermined vocabulary of tokens can map to discrete locations in the embedding space that are associated with those tokens. Other elements can be continuously distributed across the embedding space. For instance, some datatypes can be broken down into continuously defined portions (e.g., image patches) that can be described using continuously distributed locations within the embedding space.

[0111] In some implementations, the expressive power of the embedding space may not be limited to meanings associated with any particular set of tokens or other building blocks. For example, a continuous embedding space can encode a spectrum of high-order information. An individual piece of information (e.g., a token) can map to a particular point in that space: for instance, a token for the word ‘‘dog’' can be projected to an embedded value that points to a particular location in the embedding space associated with canine-related information. Similarly, an image patch of an image of a dog on grass can also be projected into the embedding space. In some implementations, the projection of the image of the dog can be similar to the projection of the word “dog’' while also having similarity to a projection of the word “grass.” while potentially being different from both. In some implementations, the projection of the image patch may not exactly align with any single projection of a single word. In some implementations, the projection of the image patch can align with a combination of the projections of the words “dog” and “grass.” In this manner, for instance, a high-order embedding space can encode information that can be independent of data modalities in which the information is expressed.

[0112] Task indicator 9 can include a model or model component configured to identify a task being performed and inject, into input sequence 8, an input value represented by element 8-0 that signals which task is being performed. For instance, the input value can be provided as a data type associated with an input modality and projected along with that input modality (e.g., the input value can be a textual task label that is embedded along with other textual datain the input; the input value can be a pixel-based representation of a task that is embedded along with other image data in the input; etc.). The input value can be provided as a data ty pe that differs from or is at least independent from other input(s). For instance, the input value represented by element 8-0 can be a learned within a continuous embedding space.

[0113] Input modalities 10-1, 10-2, and 10-3 can be associated with various different data types (e.g., as described above with respect to input(s) 2 and output(s) 3).

[0114] Data-to-sequence models 11-1, 11-2, and 11-3 can be the same or different from each other. Data-to-sequence models 11-1, 11-2, and 11-3 can be adapted to each respective input modality710-1, 10-2, and 10-3. For example, a textual data-to-sequence model can subdivide a portion of input text and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-1, 8-2, 8-3. etc.). An image data-to-sequence model can subdivide an input image and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-4, 8-5, 8-6, etc.). An arbitrary datatype data-to-sequence model can subdivide an input of that arbitrary datatype and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-7. 8-8, 8-9, etc.).

[0115] Data-to-sequence models 11-1. 11-2, and 11-3 can form part of machine-learned sequence processing model(s) 4. Data-to-sequence models 11-1, 11-2, and 11-3 can be jointly trained with or trained independently from machine-learned sequence processing model(s) 4. Data-to-sequence models 11-1, 11-2, and 11-3 can be trained end-to-end with machine-learned sequence processing model(s) 4.Example Machine-Learned Model Development Platform

[0116] FIG. 12 is a block diagram of an example model development platform 12 that can facilitate creation, adaptation, and refinement of example machine-learned models (e.g., machine-learned model(s) 1, sequence processing model(s) 4, etc.). Model development platform 12 can provide a number of different toolkits that developer systems can employ in the development of new or adapted machine-learned models.

[0117] Model development platform 12 can provide one or more model libraries 13 containing building blocks for new models. Model libraries 13 can include one or more pretrained foundational models 13-1, which can provide a backbone of processing power across various tasks. Model libraries 13 can include one or more pre-trained expert models 13-2, which can be focused on performance in particular domains of expertise. Model libraries 13 can include various model primitives 13-3, which can provide low-level architectures orcomponents (optionally pre-trained), which can be assembled in various arrangements as desired.

[0118] Model development platform 12 can receive selections of various model components 14. Model development platform 12 can pass selected model components 14 to a workbench 15 that combines selected model components 14 into a development model 16.

[0119] Workbench 15 can facilitate further refinement and adaptation of development model 16 by leveraging a number of different toolkits integrated with model development platform 12. For example, workbench 15 can facilitate alignment of the development model 16 with a desired performance profile on various tasks using a model alignment toolkit 17.

[0120] Model alignment toolkit 17 can provide a number of tools for causing development model 16 to generate outputs aligned with desired behavioral characteristics. Alignment can include increasing an accuracy, precision, recall, etc. of model outputs.Alignment can include enforcing output styles, schema, or other preferential characteristics of model outputs. Alignment can be general or domain-specific. For instance, a pre- trained foundational model 13-1 can begin with an initial level of performance across multiple domains. Alignment of the pre-trained foundational model 13-1 can include improving a performance in a particular domain of information or tasks (e g., even at the expense of performance in another domain of information or tasks).

[0121] Model alignment toolkit 17 can integrate one or more dataset(s) 17-1 for aligning development model 16. Curated dataset(s) 17-1 can include labeled or unlabeled training data. Dataset(s) 17-1 can be obtained from public domain datasets. Dataset(s) 17-1 can be obtained from private datasets associated with one or more developer system(s) for the alignment of bespoke machine-learned model(s) customized for private use-cases.

[0122] Pre-training pipelines 17-2 can include a machine-learned model training workflow configured to update development model 16 over large-scale, potentially noisy datasets. For example, pre-training can leverage unsupervised learning techniques (e.g., denoising, etc.) to process large numbers of training instances to update model parameters from an initialized state and achieve a desired baseline performance. Pre- training pipelines 17-2 can leverage unlabeled datasets in dataset(s) 17-1 to perform pre-training. Workbench 15 can implement a pre-training pipeline 17-2 to pre-train development model 16.

[0123] Fine-tuning pipelines 17-3 can include a machine-learned model training workflow configured to refine the model parameters of development model 16 with higher-quality data. Fine-tuning pipelines 17-3 can update development model 16 by conducting supervised training with labeled dataset(s) in dataset(s) 17-1. Fine-tuning pipelines 17-3 canupdate development model 16 by conducting reinforcement learning using reward signals from user feedback signals. Workbench 15 can implement a fine-tuning pipeline 17-3 to finetune development model 16.

[0124] Prompt libraries 17-4 can include sets of inputs configured to induce behavior aligned with desired performance criteria. Prompt libraries 17-4 can include few-shot prompts (e.g., inputs providing examples of desired model outputs for prepending to a desired runtime query), chain-of-thought prompts (e.g., inputs providing step-by-step reasoning within the exemplars to facilitate thorough reasoning by the model), and the like.

[0125] Example prompts can be retrieved from an available repository of prompt libraries 17-4. Example prompts can be contributed by one or more developer systems using workbench 15.

[0126] In some implementations, pre-trained or fine-tuned models can achieve satisfactory performance without exemplars in the inputs. For instance, zero-shot prompts can include inputs that lack exemplars. Zero-shot prompts can be within a domain within a training dataset or outside of the training domain(s).

[0127] Prompt libraries 17-4 can include one or more prompt engineering tools. Prompt engineering tools can provide workflows for retrieving or learning optimized prompt values. Prompt engineering tools can facilitate directly learning prompt values (e.g., input element values) based one or more training iterations. Workbench 15 can implement prompt engineering tools in development model 16.

[0128] Prompt libraries 17-4 can include pipelines for prompt generation. For example, inputs can be generated using development model 16 itself or other machine-learned models. In this manner, for instance, a first model can process information about a task and output a input for a second model to process in order to perform a step of the task. The second model can be the same as or different from the first model. Workbench 15 can implement prompt generation pipelines in development model 16.

[0129] Prompt libraries 17-4 can include pipelines for context injection. For instance, a performance of development model 16 on a particular task can improve if provided with additional context for performing the task. Prompt libraries 17-4 can include software components configured to identify desired context, retrieve the context from an external source (e.g., a database, a sensor, etc.), and add the context to the input prompt. Workbench 15 can implement context injection pipelines in development model 16.

[0130] Although various training examples described herein with respect to model development platform 12 refer to “pre-training” and “fine-tuning,” it is to be understood thatmodel alignment toolkit 17 can generally support a wide variety of training techniques adapted for training a wide variety of machine-learned models. Example training techniques can correspond to the example training method 700 described above.

[0131] Model development platform 12 can include a model plugin toolkit 18. Model plugin toolkit 18 can include a variety of tools configured for augmenting the functionality of a machine-learned model by integrating the machine-learned model with other systems, devices, and software components. For instance, a machine-learned model can use tools to increase performance quality where appropriate. For instance, deterministic tasks can be offloaded to dedicated tools in lieu of probabilistically performing the task with an increased risk of error. For instance, instead of autoregressively predicting the solution to a system of equations, a machine-learned model can recognize a tool to call for obtaining the solution and pass the system of equations to the appropriate tool. The tool can be a traditional system of equations solver that can operate deterministically to resolve the system of equations. The output of the tool can be returned in response to the original query'. In this manner, tool use can allow some example models to focus on the strengths of machine-learned models — e.g., understanding an intent in an unstructured request for a task — while augmenting the performance of the model by offloading certain tasks to a more focused tool for rote application of deterministic algorithms to a well-defined problem.

[0132] Model plugin toolkit 18 can include validation tools 18-1. Validation tools 18-1 can include tools that can parse and confirm output(s) of a machine-learned model.Validation tools 18-1 can include engineered heuristics that establish certain thresholds applied to model outputs. For example, validation tools 18-1 can ground the outputs of machine-learned models to structured data sources (e.g., to mitigate ‘‘hallucinations’').

[0133] Model plugin toolkit 18 can include tooling packages 18-2 for implementing one or more tools that can include scripts or other executable code that can be executed alongside development model 16. Tooling packages 18-2 can include one or more inputs configured to cause machine-learned model (s) to implement the tools (e.g., few-shot prompts that induce a model to output tool calls in the proper syntax, etc.). Tooling packages 18-2 can include, for instance, fine-tuning training data for training a model to use a tool.

[0134] Model plugin toolkit 18 can include interfaces for calling external application programming interfaces (APIs) 18-3. For instance, in addition to or in lieu of implementing tool calls or tool code directly with development model 16, development model 16 can be aligned to output instruction that initiate API calls to send or obtain data via external systems.

[0135] Model plugin toolkit 18 can integrate with prompt libraries 17-4 to build a catalog of available tools for use with development model 16. For instance, a model can receive, in an input, a catalog of available tools, and the model can generate an output that selects a tool from the available tools and initiates a tool call for using the tool.

[0136] Model development platform 12 can include a computational optimization toolkit 19 for optimizing a computational performance of development model 16. For instance, tools for model compression 19-1 can allow development model 16 to be reduced in size while maintaining a desired level of performance. For instance, model compression 19-1 can include quantization workflows, weight pruning and sparsification techniques, etc. Tools for hardware acceleration 19-2 can facilitate the configuration of the model storage and execution formats to operate optimally on different hardware resources. For instance, hardware acceleration 19-2 can include tools for optimally sharding models for distributed processing over multiple processing units for increased bandwidth, lower unified memory requirements, etc. Tools for distillation 19-3 can provide for the training of lighter-weight models based on the knowledge encoded in development model 16. For instance, development model 16 can be a highly performant, large machine-learned model optimized using model development platform 12. To obtain a lightweight model for running in resource-constrained environments, a smaller model can be a “student model” that learns to imitate development model 16 as a “teacher model.” In this manner, for instance, the investment in learning the parameters and configurations of development model 16 can be efficiently transferred to a smaller model for more efficient inference.

[0137] Workbench 15 can implement one, multiple, or none of the toolkits implemented in model development platform 12. Workbench 15 can output an output model 20 based on development model 16. Output model 20 can be a deployment version of development model 16. Output model 20 can be a development or training checkpoint of development model 16. Output model 20 can be a distilled, compressed, or otherwise optimized version of development model 16.

[0138] FIG. 13 is a block diagram of an example training flow for training a machine-learned development model 16. One or more portion(s) of the example training flow can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of the example training flow can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of the example training flow can be implemented on the hardware components of the device(s)described herein, for example, to train one or more systems or models. FIG. 13 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. FIG. 13 is described with reference to elements / terms described with respect to other systems and figures for exemplars' illustrated purposes and is not meant to be limiting. One or more portions of the example training flow can be performed additionally, or alternatively, by other systems.

[0139] Initially, development model 16 can persist in an initial state as an initialized model 21. Development model 16 can be initialized with weight values. Initial weight values can be random or based on an initialization schema. Initial weight values can be based on prior pre-training for the same or for a different model.

[0140] Initialized model 21 can undergo pre-training in a pre-training stage 22. Pretraining stage 22 can be implemented using one or more pre-training pipelines 17-2 over data from dataset(s) 17-1. Pre-training can be omitted, for example, if initialized model 21 is already pre-trained (e g., development model 16 contains, is, or is based on a pre-trained foundational model or an expert model).

[0141] Pre-trained model 23 can then be a new version of development model 16, which can persist as development model 16 or as anew development model. Pre-trained model 23 can be the initial state if development model 16 was already pre-trained. Pre-trained model 23 can undergo fine-tuning in a fine-tuning stage 24. Fine-tuning stage 24 can be implemented using one or more fine-tuning pipelines 17-3 over data from dataset(s) 17-1. Fine-tuning can be omitted, for example, if a pre-trained model as satisfactory performance, if the model was already fine-tuned, or if other tuning approaches are preferred.

[0142] Fine-tuned model 29 can then be anew version of development model 16, which can persist as development model 16 or as anew development model. Fine-tuned model 29 can be the initial state if development model 16 was already fine-tuned. Fine-tuned model 29 can undergo refinement with user feedback 26. For instance, refinement with user feedback 26 can include reinforcement learning, optionally based on human feedback from human users of fine-tuned model 25. As reinforcement learning can be a form of fine-tuning, it is to be understood that fine-tuning stage 24 can subsume the stage for refining with user feedback 26. Refinement with user feedback 26 can produce a refined model 27. Refined model 27 can be output to downstream system(s) 28 for deployment or further development.

[0143] In some implementations, computational optimization operations can be applied before, during, or after each stage. For instance, initialized model 21 can undergo computational optimization 29-1 (e.g., using computational optimization toolkit 19) before pre-training stage 22. Pre-trained model 23 can undergo computational optimization 29-2 (e.g., using computational optimization toolkit 19) before fine-tuning stage 24. Fine-tuned model 25 can undergo computational optimization 29-3 (e.g., using computational optimization toolkit 19) before refinement with user feedback 26. Refined model 27 can undergo computational optimization 29-4 (e.g., using computational optimization toolkit 19) before output to downstream system(s) 28. Computational optimization(s) 29-1, . . . , 29-4 can all be the same, all be different, or include at least some different optimization techniques.Example Machine-Learned Model Inference System

[0144] FIG. 14 is a block diagram of an inference system for operating one or more machine-learned model(s) 1 to perform inference (e.g., for training, for deployment, etc.). A model host 31 can receive machine-learned model(s) 1. Model host 31 can host one or more model instance(s) 31-1, which can be one or multiple instances of one or multiple models. Model host 31 can host model instance(s) 31-1 using available compute resources 31-2 associated with model host 31.

[0145] Model host 31 can perform inference on behalf of one or more client(s) 32.Client(s) 32 can transmit an input request 33 to model host 31. Using input request 33, model host 31 can obtain input(s) 2 for input to machine-learned model(s) 1. Machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3. Using output(s) 3, model host 31 can return an output payload 34 for responding to input request 33 from client(s) 32. Output payload 34 can include or be based on output(s) 3.

[0146] Model host 31 can leverage various other resources and tools to augment the inference task. For instance, model host 31 can communicate with tool interfaces 35 to facilitate tool use by model instance(s) 31-1. Tool interfaces 35 can include local or remote APIs. Tool interfaces 35 can include integrated scripts or other software functionality. Model host 31 can engage online learning interface(s) 36 to facilitate ongoing improvements to machine-learned model(s) 1. For instance, online learning interface(s) 36 can be used within reinforcement learning loops to retrieve user feedback on inferences served by model host 31. Model host 31 can access runtime data source(s) 37 for augmenting input(s) 2 with additional contextual information. For instance, runtime data source(s) 37 can include a knowledgegraph 37-1 that facilitates structured information retrieval for information associated with input request(s) 33 (e.g., a search engine service). Runtime data source(s) 37 can include public or private, external or local database(s) 37-2 that can store information associated with input request(s) 33 for augmenting input(s) 2. Runtime data source(s) 37 can include account data 37-3 which can be retrieved in association with a user account corresponding to a client 32 for customizing the behavior of model host 31 accordingly.

[0147] Model host 31 can be implemented by one or multiple computing devices or systems. Client(s) can be implemented by one or multiple computing devices or systems, which can include computing devices or systems shared with model host 31.

[0148] For example, model host 31 can operate on a server sy stem that provides a machine-learning service to client device(s) that operate client(s) 32 (e.g., over a local or wide-area network). Client device(s) can be end-user devices used by individuals. Client device(s) can be server systems that operate client(s) 32 to provide various functionality as a senice to downstream end-user devices.

[0149] In some implementations, model host 31 can operate on a same device or system as client(s) 32. Model host 31 can be a machine-learning service that runs on-device to provide machine-learning functionality to one or multiple applications operating on a client device, which can include an application implementing client(s) 32. Model host 31 can be a part of a same application as client(s) 32. For instance, model host 31 can be a subroutine or method implemented by one part of an application, and client(s) 32 can be another subroutine or method that engages model host 31 to perform inference functions within the application. It is to be understood that model host 31 and client(s) 32 can have various different configurations.

[0150] Model instance(s) 31-1 can include one or more machine-learned models that are available for performing inference. Model instance(s) 31-1 can include w eights or other model components that are stored on in persistent storage, temporarily cached, or loaded into high-speed memoiy. Model instance(s) 31-1 can include multiple instance(s) of the same model (e.g., for parallel execution of more requests on the same model). Model instance(s) 31-1 can include instance(s) of different model(s). Model instance(s) 31-1 can include cached intermediate states of active or inactive model(s) used to accelerate inference of those models. For instance, an inference session with a particular model may generate significant amounts of computational results that can be re-used for future inference runs (e.g., using a KV cache for transformer-based models). These computational results can be saved inassociation with that inference session so that session can be executed more efficiently when resumed.

[0151] Compute resource(s) 31-2 can include one or more processors (central processing units, graphical processing units, tensor processing units, machine-learning accelerators, etc.) connected to one or more memory devices. Compute resource(s) 31-2 can include a dynamic pool of available resources shared with other processes. Compute resource(s) 31-2 can include memory devices large enough to fit an entire model instance in a single memory instance. Compute resource(s) 31-2 can also shard model instance(s) across multiple memory devices (e.g., using data parallelization or tensor parallelization, etc.). This can be done to increase parallelization or to execute a large model using multiple memory devices which individually might not be able to fit the entire model into memory.

[0152] Input request 33 can include data for input(s) 2. Model host 31 can process input request 33 to obtain input(s) 2. Input(s) 2 can be obtained directly from input request 33 or can be retrieved using input request 33. Input request 33 can be submitted to model host 31 via an API.

[0153] Model host 31 can perform inference over batches of input requests 33 in parallel. For instance, a model instance 31-1 can be configured with an input structure that has a batch dimension. Separate input(s) 2 can be distributed across the batch dimension (e.g., rows of an array). The separate input(s) 2 can include completely different contexts. The separate input(s) 2 can be multiple inference steps of the same task. The separate input(s) 2 can be staggered in an input structure, such that any given inference cycle can be operating on different portions of the respective input(s) 2. In this manner, for instance, model host 31 can perform inference on the batch in parallel, such that output(s) 3 can also contain the batch dimension and return the inference results for the batched input(s) 2 in parallel. In this manner, for instance, batches of input request(s) 33 can be processed in parallel for higher throughput of output payload(s) 34.

[0154] Output payload 34 can include or be based on output(s) 3 from machine-learned model(s) 1. Model host 31 can process output(s) 3 to obtain output payload 34. This can include chaining multiple rounds of inference (e.g., iteratively, recursively, across the same model(s) or different model(s)) to arrive at a final output for a task to be returned in output payload 34. Output payload 34 can be transmitted to client(s) 32 via an API.

[0155] Online learning interface(s) 36 can facilitate reinforcement learning of machine-learned model(s) 1. Online learning interface(s) 36 can facilitate reinforcement learning withhuman feedback (RLHF). Online learning interface(s) 36 can facilitate federated learning of machine-learned model(s) 1.

[0156] Model host 31 can execute machine-learned model(s) 1 to perform inference for various tasks using various types of data. For example, various different input(s) 2 and output(s) 3 can be used for various different tasks. In some implementations, input(s) 2 can be or otherwise represent image data. Machine-learned model(s) 1 can process the image data to generate an output. As an example, machine-learned model(s) 1 can process the image data to generate an image recognition output (e.g., a recognition of the image data, a latent embedding of the image data, an encoded representation of the image data, a hash of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an image segmentation output. As another example, machine-learned model(s) 1 can process the image data to generate an image classification output. As another example, machine-learned model(s) 1 can process the image data to generate an image data modification output (e.g., an alteration of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an encoded image data output (e.g., an encoded and / or compressed representation of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an upscaled image data output. As another example, machine-learned model(s) 1 can process the image data to generate a prediction output.

[0157] In some implementations, the task is a computer vision task. In some cases. input(s) 2 includes pixel data for one or more images and the task is an image processing task. For example, the image processing task can be image classification, where the output is a set of scores, each score corresponding to a different object class and representing the likelihood that the one or more images depict an object belonging to the object class. The image processing task may be object detection, where the image processing output identifies one or more regions in the one or more images and, for each region, a likelihood that region depicts an object of interest. As another example, the image processing task can be image segmentation, where the image processing output defines, for each pixel in the one or more images, a respective likelihood for each category in a predetermined set of categories. For example, the set of categories can be foreground and background. As another example, the set of categories can be object classes. As another example, the image processing task can be depth estimation, where the image processing output defines, for each pixel in the one or more images, a respective depth value. As another example, the image processing task can be motion estimation, where the network input includes multiple images, and the imageprocessing output defines, for each pixel of one of the input images, a motion of the scene depicted at the pixel between the images in the network input.

[0158] In some implementations, input(s) 2 can be or otherwise represent natural language data. Machine-learned model(s) 1 can process the natural language data to generate an output. As an example, machine-learned model(s) 1 can process the natural language data to generate a language encoding output. As another example, machine-learned model(s) 1 can process the natural language data to generate a latent text embedding output. As another example, machine-learned model(s) 1 can process the natural language data to generate a translation output. As another example, machine-learned model(s) 1 can process the natural language data to generate a classification output. As another example, machine-learned model(s) 1 can process the natural language data to generate a textual segmentation output. As another example, machine-learned model(s) 1 can process the natural language data to generate a semantic intent output. As another example, machine-learned model(s) 1 can process the natural language data to generate an upscaled text or natural language output (e.g., text or natural language data that is higher quality than the input text or natural language, etc.). As another example, machine-learned model(s) 1 can process the natural language data to generate a prediction output (e.g., one or more predicted next portions of natural language content).

[0159] In some implementations, input(s) 2 can be or otherwise represent speech data (e.g.. data describing spoken natural language, such as audio data, textual data. etc.).Machine-learned model(s) 1 can process the speech data to generate an output. As an example, machine-learned model(s) 1 can process the speech data to generate a speech recognition output. As another example, machine-learned model(s) 1 can process the speech data to generate a speech translation output. As another example, machine-learned model(s) 1 can process the speech data to generate a latent embedding output. As another example, machine-learned model(s) 1 can process the speech data to generate an encoded speech output (e.g., an encoded and / or compressed representation of the speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate an upscaled speech output (e.g.. speech data that is higher quality than the input speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate a textual representation output (e.g., a textual representation of the input speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate a prediction output.

[0160] In some implementations, input(s) 2 can be or otherwise represent latent encoding data (e.g., a latent space representation of an input, etc.). Machine-learned model(s) 1 can process the latent encoding data to generate an output. As an example, machine-learned model(s) 1 can process the latent encoding data to generate a recognition output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a reconstruction output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a search output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a reclustering output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a prediction output.

[0161] In some implementations, input(s) 2 can be or otherwise represent statistical data. Statistical data can be, represent, or otherwise include data computed and / or calculated from some other data source. Machine-learned model(s) 1 can process the statistical data to generate an output. As an example, machine-learned model(s) 1 can process the statistical data to generate a recognition output. As another example, machine-learned model(s) 1 can process the statistical data to generate a prediction output. As another example, machine-learned model(s) 1 can process the statistical data to generate a classification output. As another example, machine-learned model(s) 1 can process the statistical data to generate a segmentation output. As another example, machine-learned model(s) 1 can process the statistical data to generate a visualization output. As another example, machine-learned model(s) 1 can process the statistical data to generate a diagnostic output.

[0162] In some implementations, input(s) 2 can be or otherwise represent sensor data. Machine-learned model(s) 1 can process the sensor data to generate an output. As an example, machine-learned model(s) 1 can process the sensor data to generate a recognition output. As another example, machine-learned model(s) 1 can process the sensor data to generate a prediction output. As another example, machine-learned model(s) 1 can process the sensor data to generate a classification output. As another example, machine-learned model(s) 1 can process the sensor data to generate a segmentation output. As another example, machine-learned model(s) 1 can process the sensor data to generate a visualization output. As another example, machine-learned model(s) 1 can process the sensor data to generate a diagnostic output. As another example, machine-learned model(s) 1 can process the sensor data to generate a detection output.

[0163] In some implementations, machine-learned model(s) 1 can be configured to perform a task that includes encoding input data for reliable and / or efficient transmission orstorage (and / or corresponding decoding). For example, the task may be an audio compression task. The input may include audio data and the output may comprise compressed audio data. In another example, the input includes visual data (e.g. one or more images or videos), the output comprises compressed visual data, and the task is a visual data compression task. In another example, the task may comprise generating an embedding for input data (e.g. input audio or visual data). In some cases, the input includes audio data representing a spoken utterance and the task is a speech recognition task. The output may comprise a text output which is mapped to the spoken utterance. In some cases, the task comprises encrypting or decry pting input data. In some cases, the task comprises a microprocessor performance task, such as branch prediction or memory' address translation.

[0164] In some implementations, the task is a generative task, and machine-learned model(s) 1 can be configured to output content generated in view of input(s) 2. For instance, input(s) 2 can be or otherwise represent data of one or more modalities that encodes context for generating additional content.

[0165] In some implementations, the task can be a text completion task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent textual data and to generate output(s) 3 that represent additional textual data that completes a textual sequence that includes input(s) 2. For instance, machine-learned model(s) 1 can be configured to generate output(s) 3 to complete a sentence, paragraph, or portion of text that follows from a portion of text represented by input(s) 2.

[0166] In some implementations, the task can be an instruction following task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent instructions to perform a function and to generate output(s) 3 that advance a goal of satisfying the instruction function (e.g., at least a step of a multi-step procedure to perform the function). Output(s) 3 can represent data of the same or of a different modality as input(s) 2. For instance, input(s) 2 can represent textual data (e.g., natural language instructions for a task to be performed) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the instructions (e.g., natural language responses, programming language responses, machine language responses, etc.). Input(s) 2 can represent image data (e.g., image-based instructions for a task to be performed, optionally accompanied by textual instructions) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the instructions (e.g., natural language responses, programming language responses, machine language responses, etc.). One or more output(s) 3 can be iteratively or recursively generated to sequentially process and accomplishsteps toward accomplishing the requested functionality. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of performing a function. Multiple steps can be performed, with a final output being obtained that is responsive to the initial instructions.

[0167] In some implementations, the task can be a question answering task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent a question to answer and to generate output(s) 3 that advance a goal of returning an answer to the question (e.g., at least a step of a multi-step procedure to perform the function). Output(s) 3 can represent data of the same or of a different modality as input(s) 2. For instance, input(s) 2 can represent textual data (e.g., natural language instructions for a task to be performed) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.). Input(s) 2 can represent image data (e.g., image-based instructions for a task to be performed, optionally accompanied by textual instructions) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.). One or more output(s) 3 can be iteratively or recursively generated to sequentially process and accomplish steps toward answering the question. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of obtaining an answer to the question (e g., querying a database, performing a computation, executing a script, etc.). Multiple steps can be performed, with a final output being obtained that is responsive to the question.

[0168] In some implementations, the task can be an image generation task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of image content. The context can include text data, image data, audio data, etc. Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent image data that depicts imagery related to the context. For instance, machine-learned model(s) 1 can be configured to generate pixel data of an image. Values for channel(s) associated with the pixels in the pixel data can be selected based on the context (e g., based on a probability determined based on the context).

[0169] In some implementations, the task can be an audio generation task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of audio content. The context can include text data, image data, audio data,etc. Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent audio data related to the context. For instance, machine-learned model(s) 1 can be configured to generate waveform data in the form of an image (e.g., a spectrogram). Values for channel(s) associated with pixels of the image can be selected based on the context. Machine-learned model(s) 1 can be configured to generate waveform data in the form of a sequence of discrete samples of a continuous waveform. Values of the sequence can be selected based on the context (e.g.. based on a probability determined based on the context).

[0170] In some implementations, the task can be a data generation task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of data (e.g., data from various data domains, such as sensor data, image data, multimodal data, statistical data. etc.). The desired data can be, for instance, synthetic data for training other machine-learned models. The context can include arbitrary data type(s).Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent data that aligns with the desired data. For instance, machine-learned model (s) 1 can be configured to generate data values for populating a dataset. Values for the data object(s) can be selected based on the context (e.g.. based on a probability determined based on the context).Example Computing Systems and Devices

[0171] FIG. 15 is a block diagram of an example networked computing system that can perform aspects of example implementations of the present disclosure. The system can include a number of computing devices and systems that are communicatively coupled over a network 49. An example computing device 50 is described to provide an example of a computing device that can perform any aspect of the present disclosure (e.g., implementing model host 31. client(s) 32, or both). An example server computing system 60 is described as an example of a server computing system that can perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). Computing device 50 and sen' er computing system(s) 60 can cooperatively interact (e.g., over network 49) to perform any aspect of the present disclosure (e.g., implementing model host 31. client(s) 32, or both). Model development platform system 70 is an example system that can host or serve model development platform(s) 12 for development of machine-learned models. Third-party system(s) 80 are example system(s) with which any of computing device 50, server computing system(s) 60, or model development platform system(s) 70 can interact in the performance of various aspects of the present disclosure (e.g., engaging third-party tools, accessing third-party databases or other resources, etc.). In an example embodiment.computing device 50 is an example of a wearable device as described herein. In an example embodiment, server computing system 60 and / or model development platform system 70 can implement a training engine as described herein.

[0172] Network 49 can be any type of communications network, such as a local area network (e.g., intranet), wide area network (e.g., Internet), or some combination thereof and can include any number of wired or wireless links. In general, communication over network 49 can be carried via any type of wired or wireless connection, using a wide variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), or protection schemes (e.g., VPN, secure HTTP, SSL). Network 49 can also be implemented via a system bus. For instance, one or more devices or systems of FIG. 15 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems.

[0173] Computing device 50 can be any type of computing device, such as, for example, a personal computing device (e.g., laptop or desktop), a mobile computing device (e.g., smartphone or tablet), a gaming console or controller, a w earable computing device, an embedded computing device, a server computing device, a virtual machine operating on a host device, or any other type of computing device. Computing device 50 can be a client computing device. Computing device 50 can be an end-user computing device. Computing device 50 can be a computing device of a service provided that provides a service to an end user (who may use another computing device to interact with computing device 50).

[0174] Computing device 50 can include one or more processors 51 and a memory 52. Processor(s) 51 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA. a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory’ 52 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory’ 52 can store data 53 and instructions 54 which can be executed by processor(s) 51 to cause computing device 50 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein.

[0175] Computing device 50 can also include one or more input components that receive user input. For example, a user input component can be a touch-sensitive component (e.g., a touch-sensitive display screen or a touch pad) that is sensitive to the touch of a user input object (e.g., a finger or a stylus). The touch-sensitive component can serve to implement avirtual keyboard. Other example user input components include a microphone, camera, LIDAR, a physical keyboard or other buttons, or other means by which a user can provide user input.

[0176] Computing device 50 can store or include one or more machine-learned models 55. Machine-learned models 55 can include one or more machine-learned model(s) 1, such as a sequence processing model 4. Machine-learned models 55 can include one or multiple model instance(s) 31-1. Machine-learned model(s) 55 can be received from server computing system(s) 60, model development platform system 70, third party system(s) 80 (e.g., an application distribution platform), or developed locally on computing device 50. Machine-learned model(s) 55 can be loaded into memory 52 and used or otherwise implemented by processor(s) 51. Computing device 50 can implement multiple parallel instances of machine-learned model(s) 55.

[0177] Server computing system(s) 60 can include one or more processors 61 and a memory762. Processor(s) 61 can be any suitable processing device (e g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality- of processors that are operatively connected. Memory' 62 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory' devices, magnetic disks, etc., and combinations thereof. Memory 62 can store data 63 and instructions 64 which can be executed by processor(s) 61 to cause server computing system(s) 60 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein.

[0178] In some implementations, server computing system 60 includes or is otherwise implemented by one or multiple server computing devices. In instances in which server computing system 60 includes multiple server computing devices, such server computing devices can operate according to sequential computing architectures, parallel computing architectures, or some combination thereof.

[0179] Server computing system 60 can store or otherwise include one or more machine-learned models 65. Machine-learned model(s) 65 can be the same as or different from machine-learned model(s) 55. Machine-learned models 65 can include one or more machine-learned model(s) 1, such as a sequence processing model 4. Machine-learned models 65 can include one or multiple model instance(s) 31-1. Machine-learned model (s) 65 can be received from computing device 50, model development platform system 70, third party system(s) 80, or developed locally on server computing system(s) 60. Machine-learned model(s) 65 can beloaded into memory 62 and used or otherw ise implemented by processor(s) 61. Server computing system(s) 60 can implement multiple parallel instances of machine-learned model(s) 65.

[0180] In an example configuration, machine-learned models 65 can be included in or otherwise stored and implemented by server computing system 60 to establish a client-server relationship with computing device 50 for serving model inferences. For instance, server computing system(s) 60 can implement model host 31 on behalf of client(s) 32 on computing device 50. For instance, machine-learned models 65 can be implemented by server computing system 60 as a portion of a web service (e.g., remote machine-learned model hosting sendee, such as an online interface for performing machine-learned model operations over a network on server computing system(s) 60). For instance, server computing system(s) 60 can communicate with computing device 50 over a local intranet or internet connection. For instance, computing device 50 can be a w orkstation or endpoint in communication with server computing system(s) 60, with implementation of machine-learned models 65 being managed by server computing system(s) 60 to remotely perform inference (e g., for runtime or training operations), with output(s) returned (e.g., cast, streamed, etc.) to computing device 50. Machine-learned models 65 can w ork cooperatively or interoperatively with machine-learned models 55 on computing device 50 to perform various tasks.

[0181] Model development platform system(s) 70 can include one or more processors 71 and a memory 72. Processor(s) 71 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 72 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 72 can store data 73 and instructions 74 which can be executed by processor(s) 71 to cause model development platform system(s) 70 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein. Example operations include the functionality described herein with respect to model development platform 12. This and other functionality can be implemented by developer tool(s) 75.

[0182] Third-party system(s) 80 can include one or more processors 81 and a memory 82. Processor(s) 81 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 82 can includeone or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 82 can store data 83 and instructions 84 which can be executed by processor(s) 81 to cause third-party system(s) 80 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein. Example operations include the functionality described herein wi th respect to tools and other external resources called when training or performing inference with machine-learned model(s) 1, 4, 16, 20, 55, 65, etc. (e.g., third-party resource(s) 85).

[0183] FIG. 15 illustrates one example arrangement of computing systems that can be used to implement the present disclosure. Other computing system configurations can be used as well. For example, in some implementations, one or both of computing system 50 or server computing system(s) 60 can implement all or a portion of the operations of model development platform system 70. For example, computing system 50 or server computing system(s) 60 can implement developer tool(s) 75 (or extensions thereof) to develop, update / train. or refine machine-learned models 1, 4. 16. 20. 55. 65. etc. using one or more techniques described herein with respect to model alignment toolkit 17. In this manner, for instance, computing system 50 or server computing system(s) 60 can develop, update / train, or refine machine-learned models based on local datasets (e.g., for model personalization / customization. as permitted by user data preference selections).

[0184] FIG. 16 is a block diagram of an example computing device 98 that performs according to example embodiments of the present disclosure. Computing device 98 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60. etc.). Computing device 98 can implement model host 31. For instance, computing device 98 can include a number of applications (e.g., applications 1 through N). Each application can contain its own machine learning library and machine-learned model(s). For example, each application can include a machine-learned model.Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. As illustrated in FIG.1 , each application can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, each application can communicate with each device component using an API (e.g., a public API). In some implementations, the API used by each application is specific to that application.

[0185] FIG. 17 is a block diagram of an example computing device 99 that performs according to example embodiments of the present disclosure. Computing device 99 can be the same as or different from computing device 98. Computing device 99 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 99 can include a number of applications (e.g., applications 1 through N). Each application can be in communication with a central intelligence layer. Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. In some implementations, each application can communicate with the central intelligence layer (and model(s) stored therein) using an API (e.g., a common API across all applications).

[0186] The central intelligence layer can include a number of machine-learned models. For example, as illustrated in FIG. 17, a respective machine-learned model can be provided for each application and managed by the central intelligence layer. In other implementations, two or more applications can share a single machine-learned model. For example, in some implementations, the central intelligence layer can provide a single model for all of the applications. In some implementations, the central intelligence layer is included within or otherwise implemented by an operating system of computing device 99.

[0187] The central intelligence layer can communicate with a central device data layer. The central device data layer can be a centralized repository of data for computing device 99. As illustrated in FIG. 17, the central device data layer can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, the central device data layer can communicate with each device component using an API (e.g., a private API).Additional Disclosure

[0188] The technology discussed herein makes reference to servers, databases, software applications, and other computer-based systems, as well as actions taken and information sent to and from such systems. The inherent flexibility of computer-based systems allows for a great variety of possible configurations, combinations, and divisions of tasks and functionality between and among components. For instance, processes discussed herein can be implemented using a single device or component or multiple devices or components working in combination. Databases and applications can be implemented on a single systemor distributed across multiple systems. Distributed components can operate sequentially or in parallel.

[0189] While the present subject matter has been described in detail with respect to various specific example embodiments thereof, each example is provided by way of explanation, not limitation of the disclosure. Those skilled in the art, upon attaining an understanding of the foregoing, can readily produce alterations to, variations of, and equivalents to such embodiments. Accordingly, the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. For instance, features illustrated or described as part of one embodiment can be used with another embodiment to yield a still further embodiment. Thus, it is intended that the present disclosure cover such alterations, variations, and equivalents.

[0190] Aspects of the disclosure have been described in terms of illustrative embodiments thereof. Any and all features in the following claims can be combined or rearranged in any way possible, including combinations of claims not explicitly enumerated in combination together, as the example claim dependencies listed herein should not be read as limiting the scope of possible combinations of features disclosed herein. Accordingly, the scope of the present disclosure is by way of example rather than by way of limitation, and the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. Moreover, terms are described herein using lists of example elements joined by conjunctions such as “and,” “or,” “but,” etc. It should be understood that such conjunctions are provided for explanatory purposes only. Clauses and other sequences of items joined by a particular conjunction such as “or,” for example, can refer to “and / or,” “at least one of’, “any combination of’ example elements listed therein, etc. Terms such as “based on” should be understood as “based at least in part on.”

[0191] The term “can” should be understood as referring to a possibility of a feature in various implementations and not as prescribing an ability that is necessarily present in every implementation. For example, the phrase “X can perform Y” should be understood as indicating that, in various implementations, X has the potential to be configured to perform Y, and not as indicating that in every instance X must always be able to perform Y. It should be understood that, in various implementations, X might be unable to perform Y and remain within the scope of the present disclosure.

[0192] The term “may” should be understood as referring to a possibility of a feature in various implementations and not as prescribing an ability that is necessarily present in every implementation. For example, the phrase “X may perform Y” should be understood as indicating that, in various implementations, X has the potential to be configured to perform Y, and not as indicating that in every instance X must always be able to perform Y. It should be understood that, in various implementations, X might be unable to perform Y and remain within the scope of the present disclosure.

Claims

WHAT IS CLAIMED IS:

1. A computer-implemented method comprising, by a computing system comprising one or more computing devices:combining audio data representing at least one noise as detected by at least one microphone of a wearable device with audio data representing user speech as detected by the at least one microphone of the wearable device;combining synthetic accelerometer data representing the at least one noise as detected by at least one accelerometer of the wearable device with synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device;providing the combined audio data and the combined synthetic accelerometer data to a machine-learned speech enhancement model configured to generate an enhanced audio output based on audio data and accelerometer data;processing the combined audio data and the combined synthetic accelerometer data to generate an output including an enhanced audio speech signal;generating a loss based on comparing the enhanced audio speech signal with the audio data representing the user speech as detected by the at least one microphone of the wearable device; andmodifying at least a portion of the machine-learned speech enhancement model based at least in part on the loss.

2. The computer-implemented method of claim 1, wherein:the machine-learned speech enhancement model includes a first transfer function that models user speech to the at least one microphone of the wearable device and second transfer function that models environmental noise to the at least one accelerometer of the wearable device.

3. The computer-implemented method of claim 1, further comprising: generating, with at least one synthetic accelerometer data generation model, the synthetic accelerometer data representing the at least one noise as detected by the at least one accelerometer of the wearable device and the synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device.

4. The computer-implemented method of claim 3, wherein:generating, with at least one synthetic accelerometer data generation model, the synthetic accelerometer data representing the at least one noise as detected by the at least one accelerometer of the wearable device comprises processing the audio data representing the at least one noise as detected by at least one microphone of the wearable device to synthesize the synthetic accelerometer data representing the at least one noise as detected by the at least one accelerometer of the wearable device; andgenerating, with at least one synthetic accelerometer data generation model, the synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device comprises processing the audio data representing the user speech as detected by the at least one microphone of the wearable device to synthesize the synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device.

5. The computer-implemented method of claim 3, wherein the at least one synthetic accelerometer data generation model includes:a first synthetic accelerometer data generation model configured to generate the synthetic accelerometer data representing the at least one noise as detected by the at least one accelerometer of the wearable device; anda second synthetic accelerometer data generation model configured to generate the synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device.6 The computer-implemented method of claim 3, wherein the at least one synthetic accelerometer data generation model is a machine-learned synthetic accelerometer data generation model.

7. The computer-implemented method of claim 1, wherein the loss is a first loss, the method further comprising:providing the enhanced audio speech signal to at least one downstream machine-learned task model;processing, with the at least one downstream machine-learned task model, the enhanced audio speech signal to generate at least one task output;generating a second loss based on comparing the at least one task output with a target task output; andmodifying at least a portion of the machine-learned speech enhancement model based at least in part on the second loss.

8. The computer-implemented method of claim 1, wherein the at least one synthetic accelerometer data generation model is a machine-learned synthetic accelerometer data generation model.

9. The computer-implemented method of claim 1, wherein the machine-learned speech enhancement model includes a symmetric encoder-decoder network.

10. A computing system, comprising:one or more processors; andone or more non-transitory computer-readable media that collectively store instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, comprising:combining audio data representing at least one noise as detected by at least one microphone of a wearable device with audio data representing user speech as detected by the at least one microphone of the wearable device;combining synthetic accelerometer data representing the at least one noise as detected by at least one accelerometer of the wearable device with synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device;providing the combined audio data and the combined synthetic accelerometer data to a machine-learned speech enhancement model configured to generate an enhanced audio output based on audio data and accelerometer data;processing the combined audio data and the combined synthetic accelerometer data to generate an output including an enhanced audio speech signal;generating a loss based on comparing the enhanced audio speech signal with the audio data representing the user speech as detected by the at least one microphone of the wearable device; andmodifying at least a portion of the machine-learned speech enhancement model based at least in part on the loss.

11. The computing system of claim 10, wherein:the machine-learned speech enhancement model includes a first transfer function that models user speech to the at least one microphone of the wearable device and second transfer function that models environmental noise to the at least one accelerometer of the wearable device.

12. The computing system of claim 11, wherein the operations further comprise: generating, with at least one synthetic accelerometer data generation model, the synthetic accelerometer data representing the at least one noise as detected by the at least one accelerometer of the wearable device and the synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device.

13. The computing system of claim 12, wherein the at least one synthetic accelerometer data generation model includes:a first synthetic accelerometer data generation model configured to generate the synthetic accelerometer data representing the at least one noise as detected by the at least one accelerometer of the wearable device; anda second synthetic accelerometer data generation model configured to generate the synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device.

14. The computing system of claim 12, wherein the at least one synthetic accelerometer data generation model is a machine-learned synthetic accelerometer data generation model.

15. The computing system of claim 10, wherein the loss is a first loss, the operations further comprising:providing the enhanced audio speech signal to at least one downstream machine-learned task model;processing, with the at least one downstream machine-learned task model, the enhanced audio speech signal to generate at least one task output;generating a second loss based on comparing the at least one task output with a target task output; andmodifying at least a portion of the machine-learned speech enhancement model based at least in part on the second loss.

16. One or more non-transitory computer-readable media that collectively store instructions that, when executed by one or more processors, cause the one or more processors to perform operations, comprising:combining audio data representing at least one noise as detected by at least one microphone of a wearable device with audio data representing user speech as detected by the at least one microphone of the wearable device;combining synthetic accelerometer data representing the at least one noise as detected by at least one accelerometer of the wearable device with synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device;providing the combined audio data and the combined synthetic accelerometer data to a machine-learned speech enhancement model configured to generate an enhanced audio output based on audio data and accelerometer data;processing the combined audio data and the combined synthetic accelerometer data to generate an output including an enhanced audio speech signal;generating a loss based on comparing the enhanced audio speech signal with the audio data representing the user speech as detected by the at least one microphone of the wearable device; andmodifying at least a portion of the machine-learned speech enhancement model based at least in part on the loss.

17. The one or more non-transitory computer-readable media of claim 10, wherein:the machine-learned speech enhancement model includes a first transfer function that models user speech to the at least one microphone of the wearable device and second transfer function that models environmental noise to the at least one accelerometer of the wearable device.

18. The one or more non-transitory computer-readable media of claim 10, wherein the operations further comprise:generating, with at least one synthetic accelerometer data generation model, the synthetic accelerometer data representing the at least one noise as detected by the at least one accelerometer of the wearable device and the synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device.

19. The one or more non-transitory computer-readable media of claim 18, wherein the at least one synthetic accelerometer data generation model includes:a first synthetic accelerometer data generation model configured to generate the synthetic accelerometer data representing the at least one noise as detected by the at least one accelerometer of the wearable device; anda second synthetic accelerometer data generation model configured to generate the synthetic accelerometer data representing the user speech as detected by the at least one accelerometer of the wearable device.

20. The one or more non-transitory computer-readable media of claim 18, wherein the at least one synthetic accelerometer data generation model is a machine-learned synthetic accelerometer data generation model.

Images