Wind noise detection and estimation in audio signals

Abstract

Systems and techniques are described herein for wind noise detection. For example, a computing device can process, using an encoder of a machine learning model, the audio signal to generate features associated with the audio signal. The computing device can process, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features. The computing device can determine, based on the classification of the audio signal, wind noise is present in the audio signal.

Description

PATENTQualcomm Reference No. 24051 0WO1WIND NOISE DETECTION AND ESTIMATION IN AUDIO SIGNALSFIELD

[0001] The present disclosure generally relates to wind noise detection and estimation. For example, aspects of the present disclosure relate to systems and techniques for wind noise detection and estimation in audio signals.BACKGROUND

[0002] Many devices use various signal detection and filtering techniques to identify and remove noise from audio signals. Noise filtering in live broadcasts can include a mix of physical and digital signal processing techniques to filter unwanted noise from an environment. Example signal filtering techniques include placing physical wind screens over microphones or applying high-pass filtering to audio signals. While physical noise filtering techniques can be passive (e.g., a windscreen), active noise filtering techniques can use computing resources that could be conserved. Many noise filtering techniques remain active even when there is not noise to be filtered. For example, many headphones can include noise-cancelling features to remove ambient noise, such as noises originating from the environment outside of the headphones, which can remain active even when there is not any noise to filter.SUMMARY

[0003] The following presents a simplified summary relating to one or more aspects disclosed herein. Thus, the following summary should not be considered an extensive overview relating to all contemplated aspects, nor should the following summary be considered to identify key or critical elements relating to all contemplated aspects or to delineate the scope associated with any particular aspect. Accordingly, the following summary has the sole purpose to present certain concepts relating to one or more aspects relating to the mechanisms disclosed herein in a simplified form to precede the detailed description presented below.

[0004] In some aspects, an apparatus for wind noise detection is provided. The apparatus includes one or more memories and one or more processors coupled to the one or more memories and configured to: process, using an encoder of a machine learning model, the audio signal to generate features associated with the audio signal; process, using a recurrent machine learningPATENTQualcomm Reference No. 24051 0WO2network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; and determine, based on the classification of the audio signal, wind noise is present in the audio signal.

[0005] In some aspects, a method for wind noise detection is provided. The method includes: processing, using an encoder of a machine learning model, the audio signal to generate features associated with the audio signal; processing, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; and determining, based on the classification of the audio signal, wind noise is present in the audio signal.

[0006] In some aspects, a non-transitory computer-readable medium is provided having stored thereon instructions that, when executed by at least one processor, cause the at least one processor to: process, using an encoder of a machine learning model, the audio signal to generate features associated with the audio signal; process, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; and determine, based on the classification of the audio signal, wind noise is present in the audio signal.

[0007] In some aspects, an apparatus for wind noise detection is provided. The apparatus includes: means for processing, using an encoder of a machine learning model, the audio signal to generate features associated with the audio signal; means for processing, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; and means for determining, based on the classification of the audio signal, wind noise is present in the audio signal.

[0008] In some aspects, one or more of the apparatuses described herein is, is part of, and / or includes a mobile device (e.g., a mobile telephone or other mobile device), an extended reality (XR) device or system (e.g., a virtual reality (VR) device, an augmented reality (AR) device, or a mixed reality (MR) device), a wearable device, a wireless communication device, a camera, a personal computer, a laptop computer, a vehicle or a computing device or component of a vehicle, a server computer or server device (e.g., an edge or cloud-based server, a personal computer acting as a server device, another device, or a combination thereof. In some aspects, the apparatus(es) canPATENTQualcomm Reference No. 24051 0WO3include a camera or multiple cameras for capturing one or more images. In some aspects, the apparatus(es) can include a display for displaying one or more images, notifications, and / or other displayable data. In some aspects, the apparatus(es) can include one or more sensors (e.g., one or more global positioning system (GPS) sensors, one or more global navigation satellite system (GNSS) sensors, one or more inertial measurement units (IMUs), such as one or more gyroscopes, one or more gyrometers, one or more accelerometers, any combination thereof, and / or other sensor).

[0009] The foregoing has outlined rather broadly the features and technical advantages of examples according to the disclosure in order that the detailed description that follows may be better understood. Additional features and advantages will be described hereinafter. The conception and specific examples disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. Such equivalent constructions do not depart from the scope of the appended claims. Characteristics of the concepts disclosed herein, both their organization and method of operation, together with associated advantages will be better understood from the following description when considered in connection with the accompanying figures. Each of the figures is provided for the purposes of illustration and description, and not as a definition of the limits of the claims. The foregoing, together with other features and aspects, will become more apparent upon referring to the following specification, claims, and accompanying drawings.

[0010] This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used in isolation to determine the scope of the claimed subject matter. The subject matter should be understood by reference to appropriate portions of the entire specification of this patent, any or all drawings, and each claim.

[0011] The preceding, together with other features and aspects, will become more apparent upon referring to the following specification, claims, and accompanying drawings.BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Illustrative aspects of the present application are described in detail below with reference to the following figures:PATENTQualcomm Reference No. 24051 0WO4

[0013] FIG. 1 illustrates an example implementation of a system-on-a-chip (SoC), in accordance with aspects of the present disclosure.

[0014] FIG. 2 illustrates example machine learning architecture for wind noise detection, in accordance with aspects of the present disclosure.

[0015] FIG. 3 illustrates an example system for wind noise detection and estimation, in accordance with aspects of the present disclosure.

[0016] FIG. 4 illustrates an example convolutional neural network block architecture for convolutional neural network for wind noise detection, in accordance with aspects of the present disclosure.

[0017] FIG. 5 illustrates an example block architecture for a one dimensional convolutional block of a machine learning model for wind noise detection, in accordance with aspects of the present disclosure.

[0018] FIG. 6 is a diagram illustrating relationships between voiced speech and wind noise of an audio signal, in accordance with aspects of the present disclosure.

[0019] FIG. 7 is a diagram illustrating example thresholding levels for wind noise detection, in accordance with aspects of the present disclosure.

[0020] FIG. 8 is a diagram illustrating example architecture of a machine learning model for wind noise detection, in accordance with aspects of the present disclosure.

[0021] FIG. 9 is a diagram illustrating example training data for training a machine learning model for wind noise detection, in accordance with aspects of the present disclosure.

[0022] FIG. 10 is a flow diagram illustrating an example process of wind noise detection, in accordance with aspects of the present disclosure.

[0023] FIG. 11 is a block diagram illustrating an example neural network, in accordance with aspects of the present disclosure.PATENTQualcomm Reference No. 24051 0WO5

[0024] FIG. 12 is a block diagram illustrating an example convolutional neural network, in accordance with aspects of the present disclosure.

[0025] FIG. 13 is a block diagram illustrating example computing device architecture of an example computing device which can implement the various techniques described herein.DETAILED DESCRIPTION

[0026] Certain aspects and embodiments of this disclosure are provided below. Some of these aspects and embodiments may be applied independently and some of them may be applied in combination as would be apparent to those of skill in the art. In the following description, for the purposes of explanation, specific details are set forth in order to provide a thorough understanding of embodiments of the application. However, it will be apparent that various embodiments may be practiced without these specific details. The figures and description are not intended to be restrictive.

[0027] The ensuing description provides example embodiments only, and is not intended to limit the scope, applicability, or configuration of the disclosure. Rather, the ensuing description of the example embodiments will provide those skilled in the art with an enabling description for implementing an example embodiment. It should be understood that various changes may be made in the function and arrangement of elements without departing from the spirit and scope of the application as set forth in the appended claims.

[0028] As mentioned, many devices and systems use various signal detection and filtering techniques to identify and remove noise from audio signals. Various physical and digital signalprocessing techniques are generally used to filter unwanted noise from an environment such as by including physical wind screens over microphones or applying high-pass filtering to audio signals. Various devices can use the physical and digital signal processing techniques to reduce noise in audio signals. For example, various headphones can include noise-cancelling features to remove ambient noise, such as noises originating from the environment outside of the headphones.

[0029] Machine learning systems (e.g., deep neural network systems or models) can be used to perform a variety of tasks in audio signal processing such as voice detection and / or voice recognition, ambient noise filtering, noise classification, among other tasks. In some cases, aPATENTQualcomm Reference No. 24051 0WO6machine learning system can be used to detect noise within an audio signal and classify the noise. For example, audio signals recorded outside can be of reduced quality due to ambient noise added to the audio signal, such as noise from wildlife, noise from traffic, noise from wind, etc. The introduction of ambient noise, such as wind noise, can make speech within an audio signal unintelligible or otherwise diminish a user’s experience listening to the audio signal.

[0030] Wind noise can occur when air movement or air turbulence, typically from wind, causes unwanted sound interference in a microphone’s audio signal. Interference from wind generally manifests as a low-frequency rumbling or a high-frequency random spikes noise. Wind noise generated by air turbulence hitting microphone membrane is often random in both time and space and may push membranes of the microphone beyond excursion points leading to microphone saturation and nonlinear distortion of the audio signal. Wind noise can significantly degrade the quality of audio recordings or live broadcasts, making it essential to use systems and methods wind noise reduction to minimize its effects for human listening. However, performance of wind noise reduction can require identifying the presence of wind noise in an audio signal as an initial step to applying wind noise reduction techniques.

[0031] Systems, apparatuses, electronic devices, methods (also referred to as processes), and computer-readable media (collectively referred to herein as “systems and techniques”) are described herein for wind noise detection. In some aspects, the systems and devices can include a machine learning model for wind noise detection. The machine learning model can use various machine learning model architectures. For example, the machine learning model can be a neural network, such as a convolutional neural network (CNN). In some examples, the machine learning model can be an end-to-end neural network. In other examples, the machine learning model can be various deep learning models such as a deep neural network (DNN), a recurrent neural network (RNN), a transformer, etc. The machine learning model can be trained on a set of training data including labeled audio data indicating the presence of wind noise in the audio data.

[0032] In some aspects, the machine learning model can receive an audio signal. For example, the machine learning model can receive an audio signal in real time or near real time (e.g., a live broadcast or livestream including an audio signal). In other examples, the machine learning model can receive audio data, such as a prerecorded audio file. The machine learning model can receivePATENTQualcomm Reference No. 24051 0WO7the audio signal (e.g., a raw audio feed) and output a probability associated with whether the audio signal includes wind noise. For example, the output of the machine learning model (or an output of a component of the machine learning model) can be a probability distribution associated with whether sections (e.g., time step sections) of the audio signal includes wind noise. The systems and techniques can include comparing the probability distribution to a predetermined wind noise threshold to determine whether wind noise is present in the audio signal or in a section of the audio signal. For example, a user or the systems and techniques can set a threshold level (e.g., a probability threshold) based on the output probability distribution, associated with whether wind noise is detected in the audio signal or in a section of the audio signal. The systems and techniques can classify the audio signal and compare the classification to the threshold level to determine whether wind noise is present in a time step section of the audio signal.

[0033] In some aspects, the machine learning model can receive an audio signal at an encoder of the machine learning model. In some examples, the encoder can include various convolutional blocks. For example, the encoder (or the machine learning model) can include a convolutional layer to process the audio signal. In some aspects, the encoder can include one or more onedimensional convolution (e.g., Convld) blocks to generate a feature map associated with the audio signal. In some examples, sections of the audio signal (e.g., time step sections) can be received as input to the encoder. For example, the time step sections can be based on a time period of the audio signal, such as 200 millisecond (ms) sections. In further examples, the time step sections can be based on a number of sampling points of the audio signal. For example, the time step sections can be based on a number of sampling points such as 3200 points.

[0034] In some aspects, the encoder can include one or more convolutional layer and one or more pooling layers. The encoder can include multiple layers (e g., convolutional, pooling, or another layer) within a block. For example, the encoder can include a convolutional block including one or more convolutional layers and other layers. The encoder can include multiple blocks. By way of a non-limiting example, the encoder can include four convolutional blocks, which can include each include one or more convolutional layers. In some examples, the convolutional blocks can include a convolutional layer (e.g., Convld), a batch normalization layer (e.g., BatchNormld), and a rectified linear unit (ReLU). The batch normalization layer canPATENTQualcomm Reference No. 24051 0WO8normalize the output of the convolutional layer to adjust the output of the convolutional layer to different dimensions (e.g., by adjusting the output of the convolutional layer to a different dimension output). The ReLU can add non-linearity to the output of the batch normalization layer to identify relationships between features of the audio signal. In some aspects, the ReLU layer can apply the function f(x) = max (0, x) to the output of the batch normalization layer to change all of the negative activations to 0 to increase non-linear properties of the convolutional block. In some examples, the output of the ReLU can be provided to a subsequent convolutional block to be further processed (e.g., the output of the ReLU received as input to another convolutional block with another convolutional layer, another batch normalization layer, and another ReLU.

[0035] In some examples, the encoder can include one or more pooling layers to reduce dimensionality of the outputs of the convolutional layers or outputs of the convolutional blocks while preserving the features of the audio signal associated with wind noise detection and wind noise estimation. The pooling layers can be used to simplify outputs of the convolutional blocks by reducing the size (e.g., the dimensions) of the outputs of the convolutional blocks. In examples, where the pooling layer include max pooling operations (e g., MaxPool Id) the pooling layer can simplify the output of the convolutional blocks while preserving prominent features (e.g., features associated with higher values of the output). The pooling layer can provide feature selection of the outputs of the convolutional blocks.

[0036] In some aspects, the output of the encoder (e.g., the encoder of the machine learning model including one or more convolutional blocks) can be a feature map associated with features of the audio signal. In some aspects, the feature map can be associated with various features of the audio signal (or time step section of the audio signal) such as frequency levels of the audio signal, signal sub-band centroids (SSC) features, pitch, decibel level, etc. For example, the feature map can be associated with features of a time step section of the audio signal. In such an example, the feature map can be associated with a 200 ms, or other predetermined time period, of the audio signal. The feature map associated with the time step section can be received by a recurrent machine learning network (e.g., an RNN) of the machine learning model. For example, the recurrent machine learning network can include two or more gated recurrent unit (GRU). In somePATENTQualcomm Reference No. 24051 0WO9examples, the recurrent machine learning network can include long short-term memory (LSTMs) instead of or in addition to the GRUs.

[0037] The recurrent machine learning network (also referred to as the recurrent machine learning model) can use an output of the recurrent machine learning network associated with a previous time step section of the audio signal to generate one or more probabilities associated with whether noise is present in a current time step section of the audio signal based on the features (e.g., features from the feature map) of the current time step section of the audio signal. In some examples, the recurrent machine learning network 1 can use half of a current time step section of the audio signal and the output of the recurrent machine learning network (e.g., when the time step section is 200 ms, the recurrent machine learning network can use outputs of the recurrent machine learning network associated with 100 ms when processing a subsequent 100 ms of the time step section) to generate one or more probabilities associated with whether noise is present in a time step section of the audio signal.

[0038] The output of the recurrent machine learning network can be provided as input to a dense layer (e.g., a fully connected layer of the neural network) to classify time step sections of the audio signal based on the probabilities output by the recurrent machine learning network. In some examples, the machine learning model can use activation functions, such as a softmax function to convert outputs of a fully connected layer of the machine learning model into probabilities associated with whether wind noise is present in a time step section of the audio signal.

[0039] In some aspects, the time step sections can be assigned levels based on a probability of whether the time step section includes wind noise. The levels can be associated with a range of wind noise values associated with a strength of the wind noise (e.g., strength in decibels) in the audio signal. For example, the systems and techniques can include determining whether the audio signal, or a time step section of the audio signal includes wind noise, and when the audio signal includes wind noise, estimating the wind noise within the audio signal. The systems and techniques can include classifying the audio signal based on a strength (e.g., in decibels) of the wind noise by assigning a level to the audio signal or to a time step section of the audio signal. In some examples, the systems and techniques can include setting a threshold level associated with the strength of wind noise to determine whether to assign a tag to the audio signal (or time step section of thePATENTQualcomm Reference No. 24051 0WO10audio signal) indicating the presence of wind noise in the audio signal or indicating no wind noise (e.g., wind noise below the predetermined threshold level of wind noise strength) in the audio signal.

[0040] In some aspects, the systems and techniques can include a sensor to determine when to perform wind noise detection. For example, the sensor can be a camera, accelerometer, or other sensor to identify when to perform wind noise detection (e.g., by detecting movement of objects in an environment due to wind such as a moving branch of a tree, etc.). In some examples, the systems and techniques can perform wind noise detection based on the location of a device receiving audio (e.g., the systems and techniques can perform wind noise detection when a device is outside).

[0041] Various aspects of the present disclosure will be described with respect to the figures.

[0042] FIG. 1 illustrates an example implementation of a system-on-a-chip (SOC) 100, which may include a central processing unit (CPU) 102 or a multi-core CPU, configured to perform one or more of the functions described herein. Parameters or variables (e.g., neural signals and synaptic weights), system parameters associated with a computational device (e.g., neural network with weights), delays, frequency bin information, task information, among other information may be stored in a memory block associated with a neural processing unit (NPU) 108, in a memory block associated with a CPU 102, in a memory block associated with a graphics processing unit (GPU) 104, in a memory block associated with a digital signal processor (DSP) 106, in a memory block 118, and / or may be distributed across multiple blocks. Instructions executed at the CPU 102 may be loaded from a program memory associated with the CPU 102 or may be loaded from a memory block 118.

[0043] The SOC 100 may also include additional processing blocks tailored to specific functions, such as a GPU 104, a DSP 106, a connectivity block 110, which may include fifth generation (5G) connectivity, fourth generation long term evolution (4G LTE) connectivity, WiFi connectivity, USB connectivity, Bluetooth connectivity, and the like, and a multimedia processor 112 that may, for example, detect and recognize audio signals. In one implementation, the NPU is implemented in the CPU 102, DSP 106, and / or GPU 104. The SOC 100 may alsoPATENTQualcomm Reference No. 24051 0WO11include one or more sensors 114 such as but not limited to one or more microphones, image signal processors (ISPs) 116, and / or storage 120.

[0044] The SOC 100 may be based on an ARM instruction set. In an aspect of the present disclosure, the instructions loaded into the CPU 102 may comprise code to search for a stored multiplication result in a lookup table (LUT) corresponding to a multiplication product of an input value and a filter weight. The instructions loaded into the CPU 102 may also comprise code to disable a multiplier during a multiplication operation of the multiplication product when a lookup table hit of the multiplication product is detected. In addition, the instructions loaded into the CPU 102 may comprise code to store a computed multiplication product of the input value and the filter weight when a lookup table miss of the multiplication product is detected.

[0045] SOC 100 and / or components thereof may be configured to perform audio processing, such as denoising audio signals of wind noise, using machine learning techniques according to aspects of the present disclosure discussed herein. For example, SOC 100 and / or components thereof may be configured to perform processing techniques such as but not limited to: segment shifting, shuffle correlation, gain shifting, segment masking, and additional processing techniques. SOC 100 can be part of a computing device or multiple computing devices. In some examples, SOC 100 can be part of an electronic device (or devices) such as an audio recording device, camera system (e.g., a digital camera, an IP camera, a video camera, a security camera, etc.), a telephone system (e.g., a smartphone, a cellular telephone, a conferencing system, etc.), a desktop computer, an XR device (e.g., a head-mounted display, etc.), a smart wearable device (e.g., a smart watch, smart glasses, etc.), a laptop or notebook computer, a tablet computer, a set-top box, a television, a display device, a system-on-chip (SoC), a digital media player, a gaming console, a video streaming device, a server, a drone, a computer in a car, an Intern et-of-Things (loT) device, or any other suitable electronic device(s).

[0046] In some implementations, the CPU 102, the GPU 104, the DSP 106, the NPU 108, the connectivity block 110, the multimedia processor 112, the one or more sensors 114, the ISPs 116, the memory block 118 and / or the storage 120 can be part of the same computing device. For example, in some cases, the CPU 102, the GPU 104, the DSP 106, the NPU 108, the connectivity block 110, the multimedia processor 112, the one or more sensors 114, the ISPs 116, the memoryPATENTQualcomm Reference No. 24051 0WO12block 118 and / or the storage 120 can be integrated into a smartphone, laptop, tablet computer, smart wearable device, video gaming system, server, and / or any other computing device. In other implementations, the CPU 102, the GPU 104, the DSP 106, theNPU 108, the connectivity block 110, the multimedia processor 112, the one or more sensors 114, the ISPs 116, the memory block 118 and / or the storage 120 can be part of two or more separate computing devices.

[0047] Machine learning (ML) can be considered a subset of artificial intelligence (Al). ML systems can include algorithms and statistical models that computer systems can use to perform various tasks by relying on patterns and inference, without the use of explicit instructions. An example of a ML system is a neural network (also referred to as an artificial neural network), which may include an interconnected group of artificial neurons (e.g., neuron models). Neural networks may be used for various applications and / or devices, such as image and / or video coding, image analysis and / or computer vision applications, Internet Protocol (IP) cameras, Internet of Things (loT) devices, autonomous vehicles, service robots, among others.

[0048] Individual nodes in a neural network may emulate biological neurons by taking input data and performing simple operations on the data. The results of the simple operations performed on the input data are selectively passed on to other neurons. Weight values are associated with each vector and node in the network, and these values constrain how input data is related to output data. For example, the input data of each node may be multiplied by a corresponding weight value, and the products may be summed. The sum of the products may be adjusted by an optional bias, and an activation function may be applied to the result, yielding the node’ s output signal or “output activation” (sometimes referred to as a feature map or an activation map). The weight values may initially be determined by an iterative flow of training data through the network (e.g., weight values are established during a training phase in which the network learns how to identify particular classes by their typical input data characteristics).

[0049] Different types of neural networks exist, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), generative adversarial networks (GANs), multilayer perceptron (MLP) neural networks, transformer neural networks, among others. For instance, convolutional neural networks (CNNs) are a type of feed-forward artificial neural network. Convolutional neural networks may include collections of artificial neurons that each have a receptive field (e.g., aPATENTQualcomm Reference No. 24051 0WO13spatially localized region of an input space) and that collectively tile an input space. RNNs work on the principle of saving the output of a layer and feeding this output back to the input to help in predicting an outcome of the layer. A GAN is a form of generative neural network that can learn patterns in input data so that the neural network model can generate new synthetic outputs that reasonably could have been from the original dataset. A GAN can include two neural networks that operate together, including a generative neural network that generates a synthesized output and a discriminative neural network that evaluates the output for authenticity. In MLP neural networks, data may be fed into an input layer, and one or more hidden layers provide levels of abstraction to the data. Predictions may then be made on an output layer based on the abstracted data.

[0050] For example, the use of recurrent connections and / or temporal information in a machine learning model for audio processing, such as denoising of audio signals, can be used to preserve low-frequency audio signals in audio signals with wind noise, to achieve higher quality audio signals. Various recurrent architectures (e.g., RNNs) that include one or more recurrent cells among the feed-forward layers of the network can be used to perform audio processing operations to generate processed output audio signals having a relatively high quality. For example, recurrent cells can be implemented using a vanilla-RNN architecture, a Conv-GRU (Gated Recurrent Unit) architecture, a Conv-LSTM (Long Short-Term Memory) architectures, among various others.

[0051] Deep learning (DL) is an example of a machine learning technique and can be considered a subset of ML. Many DL approaches are based on a neural network, such as an RNN or a CNN, and utilize multiple layers. The use of multiple layers in deep neural networks can permit progressively higher-level features to be extracted from a given input of raw data. For example, the output of a first layer of artificial neurons becomes an input to a second layer of artificial neurons, the output of a second layer of artificial neurons becomes an input to a third layer of artificial neurons, and so on. Layers that are located between the input and output of the overall deep neural network are often referred to as hidden layers. The hidden layers learn (e.g., are trained) to transform an intermediate input from a preceding layer into a slightly more abstract and composite representation that can be provided to a subsequent layer, until a final or desired representation is obtained as the final output of the deep neural network.PATENTQualcomm Reference No. 24051 0WO14

[0052] As noted above, a neural network is an example of a machine learning system, and can include an input layer, one or more hidden layers, and an output layer. Data is provided from input nodes of the input layer, processing is performed by hidden nodes of the one or more hidden layers, and an output is produced through output nodes of the output layer. Deep learning networks typically include multiple hidden layers. Each layer of the neural network can include feature maps or activation maps that can include artificial neurons (or nodes). A feature map can include a filter, a kernel, or the like. The nodes can include one or more weights used to indicate an importance of the nodes of one or more of the layers. In some cases, a deep learning network can have a series of many hidden layers, with early layers being used to determine simple and low-level characteristics of an input, and later layers building up a hierarchy of more complex and abstract characteristics.

[0053] A deep learning architecture may learn a hierarchy of features. If presented with visual data, for example, the first layer may learn to recognize relatively simple features, such as edges, in the input stream. In another example, if presented with auditory data, the first layer may learn to recognize spectral power in specific frequencies. The second layer, taking the output of the first layer as input, may learn to recognize combinations of features, such as simple shapes for visual data or combinations of sounds for auditory data. For instance, higher layers may learn to represent complex shapes in visual data or words in auditory data. Still higher layers may learn to recognize common visual objects or spoken phrases. Deep learning architectures may perform especially well when applied to problems that have a natural hierarchical structure. For example, the classification of audio can benefit from first learning to recognize individual spoken words, instruments in music, etc. These features may be combined at higher layers in different ways to recognize sounds such as speech, instruments, wind noise, etc.

[0054] Neural networks may be designed with a variety of connectivity patterns. In feed-forward networks, information is passed from lower to higher layers, with each neuron in a given layer communicating to neurons in higher layers. A hierarchical representation may be built up in successive layers of a feed-forward network, as described above. Neural networks may also have recurrent or feedback (also called top-down) connections. In a recurrent connection, the output from a neuron in a given layer may be communicated to another neuron in the same layer. APATENTQualcomm Reference No. 24051 0WO15recurrent architecture may be helpful in recognizing patterns that span more than one of the input data chunks that are delivered to the neural network in a sequence. A connection from a neuron in a given layer to a neuron in a lower layer is called a feedback (or top-down) connection. A network with many feedback connections may be helpful when the recognition of a high-level concept may aid in discriminating the particular low-level features of an input. Further description of machine learning model architecture is provided in the description of FIG. 11 and FIG. 12.

[0055] The SOC 100 of FIG. 1 can be used to perform the various systems and techniques described in the descriptions of FIG. 2-FIG. 10. For example, the SOC 100 can be used to perform operations of the machine learning architecture 200 of FIG. 2 including wind noise detection and wind noise estimation.

[0056] FIG. 2 is a block diagram illustrating an example machine learning model architecture 200 for wind noise detection and estimation. The machine learning model architecture 200 includes an encoder 204, a recurrent machine learning network 208, and a dense layer 210 (e.g., a classification model). The machine learning model architecture 200 can be an end-to-end machine learning model, such as an end-to-end neural network. The machine learning architecture can include a convolutional neural network, a deep neural network, a recurrent neural network, transformer architecture, etc. For example, the encoder 204 can include a convolutional neural network (CNN), which in some cases can include a plurality of one-dimensional (ID) CNN blocks..

[0057] The encoder 204 can receive an audio signal 202 as input. In some examples, the encoder 204 can partition (e.g., divide) the audio signal 202 into time step sections. For example, the encoder 204 can receive the audio signal 202 in entirety (e.g., as an audio file) and divide the audio signal 202 into predetermined time step sections, such as dividing the audio signal into a sequence of 200 ms sections. In other examples, the audio signal 202 can be a signal received by the encoder 204 in real-time (e.g., audio signal during a live broadcast or when a device is actively recording audio) and the encoder 204 can receive as input the audio signal into predetermined time step sections. In further examples, the encoder 204 can partition the audio signal into sections based on a number of samples (e.g., 3200 samples per time step section). Further description of example architecture of the encoder 204 is provided in the description of FIG. 4 and FIG. 5.PATENTQualcomm Reference No. 24051 0WO16

[0058] The encoder 204 can output a feature map 206 associated with features of the audio signal 202. In some examples, the feature map 206 is associated with features of a time step section of the audio signal (e.g., features associated with a 200 ms section of the audio signal). In further examples, the feature map can be associated with a predetermined number of samples associated with the audio signal 202 (e.g., 3200 samples). In one illustrative example, the feature map can be a feature vector (or tensor), such as a ID feature vector (e.g., having a dimension of 128 x 1) in cases when the encoder includes ID CNN blocks.

[0059] The encoder 204 can output a feature map associated with features of the audio signal. In some aspects, the feature map can be associated with various features of the audio signal 202 (or time step section of the audio signal) such as frequency levels of the audio signal, signal subband centroids (SSC) features, pitch, decibel level, etc. For example, the feature map can be associated with features of a time step section of the audio signal 202. In such an example, the feature map 206 can be associated with a predetermined time period of the audio signal 202. The encoder 204 can generate additional feature maps associated with subsequent time step sections of the audio signal 202.

[0060] The feature map 206 can be received as input to the recurrent machine learning network 208. The recurrent machine learning network 208 can include one or more gated recurrent unit (GRU) blocks (e.g., with one illustrative GRU block shown in FIG. 2). In further examples, other machine learning architectures can be used in the recurrent machine learning network 208, such as long short-term memory (LSTMs). The recurrent machine learning network 208 receive as input the feature map 206 and an output probability associated with a previous time step section of the audio signal 202. For example, the recurrent machine learning network 208 can receive as input the feature map 206 and a probability associated with a previous time step section of the audio signal to output a probability associated with whether wind noise is present in a current time step section of the audio signal 202. The probability associated with the current time step section can be used as input for a subsequent time step section.

[0061] In some examples, the recurrent machine learning network 208 can use a portion of a feature map associated with a portion of the time step section as input to the recurrent machine learning network 208. For example, the recurrent machine learning network 208 can use half (orPATENTQualcomm Reference No. 24051 0WO17substantially half) of a current time step section of the audio signal and a previous output of the recurrent machine learning network associated with the other half of the time step section to generate a probability associated with the time step section of the audio signal (e.g., when the time step section is 200 ms, the recurrent machine learning network can use outputs of the recurrent machine learning network associated with a first 100 ms when processing a subsequent 100 ms of the time step section).

[0062] The output of the recurrent machine learning network 208 (e g., probabilities associated with whether wind noise is present in a time step section of the audio signal or within the audio signal in entirety) can be received as input to a dense layer 210 of the machine learning architecture 200. The dense layer 210 can be a fully connected layer of the machine learning architecture (e.g., when the machine learning architecture is a neural network, the dense layer 210 can be a fully connected layer of the neural network).

[0063] The dense layer 210 classify the outputs of the recurrent machine learning network 208 to levels associated with wind noise. In some examples, the machine learning model can use activation functions, such as a softmax function to convert outputs of a fully connected layer of the machine learning model into probabilities associated with whether wind noise is present in a time step section of the audio signal. For example, the dense layer 210 can classify the outputs of the recurrent machine learning network 208 by assigning a wind noise level associated with time step sections of the audio signal 202. In some examples, the dense layer 210 can assign a tag (e.g., a label) to the time step sections of the audio signal 202 indicating whether the wind noise is present or not present (e.g., a binary indication of whether wind noise is present).

[0064] In some examples, the dense layer 210 can assign time step sections a level (e.g., bins or level bins) based on a probability of whether the time step section includes wind noise. The levels can correspond to a range of wind noise values associated with a strength of the wind noise (e.g., strength in decibels) in the audio signal. In some examples, the machine learning architecture 200 can include a thresholding engine (not shown) to set a threshold level associated with when a time step section is determined to include wind noise. In further examples, the thresholding engine, or the dense layer 210, can set ranges of wind noise based on the strength of wind noise (e g., wind noise represented as wind noise values associated with strength in decibels, etc.) corresponding toPATENTQualcomm Reference No. 24051 0WO18the levels. For example, wind noise from 0 to 20 decibels can be associated with a first level, and wind noise from 20 to 30 decibels can be associated with a second level, etc.

[0065] FIG. 3 is an example system 300 for performing wind noise detection and wind noise estimation. System 300 can include a machine learning model 304 and a thresholding engine 306. In some examples, the thresholding engine 306 is part of the machine learning model 304. For example, the dense layer 210 of FIG. 2 can perform thresholding of wind noises (e.g., determining the ranges of wind noise associated with various levels of wind noise estimation).

[0066] As described in the description of the machine learning architecture 200 of FIG. 2, the machine learning model 304 can receive as input an audio signal 302. The machine learning model 304 can process the audio signal 302 to determine features associated with the audio signal 302 and probabilities associated with whether the audio signal 302 includes wind noise. The machine learning model 304 can output the probabilities and features associated with the audio signal 302 to a thresholding engine. The thresholding engine 306 can set ranges corresponding with different levels of wind noise in the audio signal 302. In some examples, the ranges and levels can be based on the amount wind noise present in previous audio signals or other time step sections of the audio signal 302. For example, the thresholding engine 306 can set wind noise value ranges associated with strength of the wind noise within an audio signal 302. In such an example, the ranges can vary based on the audio signal.

[0067] The thresholding engine 306 can output levels associated with time step sections of the audio signal 302, a level associated with the entirety of the audio signal 302, or a tag indicating the presence of wind noise (or a lack of wind noise) in the audio signal 302. The output level can be used by downstream applications 308 or processes, such as to activate wind noise reduction applications and techniques. For example, the downstream applications 308 can include a wind noise activity indicator to indicate a binary determination of whether wind noise is present in the audio signal 302 (e g., 1= YES WIND NOISE PRESENT, 2=NO WIND NOISE PRESENT).

[0068] FIG. 4 is a block diagram illustrating an example encoder 400, such as the encoder 204 of FIG. 2, for wind noise detection and wind noise estimation. The encoder 400 can include one or more convolutional blocks (e.g., a first convolutional block 402 and a subsequent convolutional block 406) and one or more pooling layers (e.g., a first pooling layer 404 and a subsequent poolingPATENTQualcomm Reference No. 24051 0WO19layer 408). Further description of convolutional blocks is provided in the description of FIG. 5. The encoder 400 can be part of a machine learning model, such as the machine learning model architecture 200 of FIG. 2, the machine learning model 304 of FIG. 3, the neural network 1100 of FIG. 11, the convolutional neural network 1200 of FIG. 12, or any other machine learning model described herein.

[0069] The encoder can include one or more convolutional layers and one or more pooling layers. Multiple layers can be defined as blocks of the encoder 400. For example, the encoder can have a plurality of convolutional blocks. Within the convolutional blocks can be one or more convolutional layers and other layers. For example, the convolutional blocks can include a convolutional layer for performing one-dimensional convolutions (e.g., Convld).

[0070] The convolutional blocks can include different kernel sizes and can receive inputs of different dimensions. For example, the first convolutional block 402 can have a kernel size of 64 (e.g., 64 samples in a one-dimensional audio signal). The first convolutional block 402 can have a stride of 2 samples (e.g., a distance in samples across which the kernel during convolutions). The first convolutional block 402 can include an input channel for receiving a predetermined dimension of inputs (e.g., N number of samples) and an output channel for outputting features of the input (e.g., 16 can represent a number of features or feature maps output by the first convolutional block 402).

[0071] The output of the first convolutional block 402 can be received as input by the first pooling layer 404. The first pooling layer 404 can downsample the output of the convolutional block to reduce dimensions of the output. For example, the first pooling layer 404 can be a one dimensional pooling layer such as MaxPool Id. The pooling layers can be used to reduce dimensionality of the outputs of the convolutional layers or outputs of the convolutional blocks while preserving prominent features (e.g., features associated with a higher value). For example, in a machine learning model trained for wind noise detection and estimation, the prominent features can be features associated with the presence of wind noise in a signal, such as features associated with frequency levels of the audio signal, signal sub-band centroids (SSC) features, pitch, decibel level, etc.PATENTQualcomm Reference No. 24051 0WO20

[0072] The subsequent convolutional block 406 can have different parameters than the first convolutional block 402 (e.g., different kernel size, stride, in channels, and out channels). For example, the encoder 400 can include a pooling layer between convolutional blocks to incrementally reduce the dimensions of inputs and outputs of the convolutional blocks. The subsequent pooling layer 408 can have different parameters including a different kernel size and stride. The subsequent pooling layer 408 can generate outputs of different dimensions than the first pooling layer 404. In some examples, the output of the subsequent pooling layer 408 can be a feature map, such as the feature map 206 of FIG. 2.

[0073] FIG. 5 is a block diagram illustrating example architecture of a convolutional block 500. The convolutional block 500 can be part of an encoder, such as the encoder 400 of FIG. 4 or the encoder 204 of FIG. 2. The convolutional block 500 can be the first convolutional block 402, the subsequent convolutional block 406, or any of the other convolutional blocks illustrated in FIG. 4.

[0074] The convolutional block 500 includes a convolutional layer 502, a batch normalization layer 504, and a rectified linear unit (ReLU) layer 506. The convolutional layer 502 can be used to perform various convolutional operations. By way of non-limiting example, the convolutional layer 502 can perform a one-dimensional convolution (e.g., Convld). The output of the convolutional layer 502 (e.g., a feature map) can be received by the batch normalization layer 504.

[0075] The batch normalization layer 504 can normalize the output of the convolutional layer. In some examples, the batch normalization layer can adjust dimensions of the output of the convolutional layer. In further examples, the batch normalization layer can scale and shift outputs of the convolutional layer (e.g., shift positions of features within a feature map output by the convolutional layer, scale features, normalize features, etc.).

[0076] The output of the batch normalization layer 504 can be received by the ReLU layer 506. The ReLU layer 506 can add non-linearity to the output of the batch normalization layer 504. The non-linearity can be used by a machine learning model to identify relationships between features of a feature map (e.g., relationships of features of an audio signal to identify wind noise). In some examples, the ReLU layer 506 can apply the function f(x) = max (0, x) to the output of the batch normalization layer 504 to change negative activations to 0 to increase non-linear properties of the convolutional block 500. In some examples, the output of the ReLU layer 506 can be provided toPATENTQualcomm Reference No. 24051 0WO21a subsequent convolutional block or a pooling layer (e.g., the pooling layers of FIG. 4) to be further processed.

[0077] FIG. 6 is a diagram 600 illustrating a relationship between voiced speech and wind noise of an audio signal based on signal sub-band centroids of an audio signal. For example, signal subband centroids can refer to central frequencies of a frequency band (e.g., sub-band) of an audio signal. When a centroid of the audio signal is less than 100 Hz, the audio signal can demonstrate a higher incidence of including wind noise. Various machine learning models and techniques, such as the machine learning architecture 200 of FIG. 2, the machine learning model 304 of FIG. 3, etc.) can be used to identify relationships between wind noise and other sounds in an audio signal, such as voiced speech. Other example features of audio signals that can indicate presence of wind noise can include pitch, decibel level, and various frequency characteristics of the audio signal.

[0078] FIG. 7 is a diagram 700 illustrating example levels wind noise estimation. For example, the diagram 700 includes a level corresponding to a range of decibel values associated with a strength of the wind noise. Each level can include a corresponding label (e.g., 0 to 7) indicating an estimation of the strength of the wind noise. In some examples, a thresholding engine such as the thresholding engine 306 of FIG. 3, can adjust the range of decibel values corresponding to the levels. For example, the diagram 700 includes a range from 0 to -8 dB corresponding to level 7. In another example, the range corresponding to level 7 can be set at 0 to -12 dB or another range of values based on the wind noise detected in the audio signal. In some examples, the wind level can be represented as different bins (e.g., discrete ranges of wind strength assigned with each range assigned a different value such as 1, 2, 3, etc.). For example, the wind level (e.g., Level_label_dB) can be represented as follows:Level_label_dB = 20 * torch. loglO(calculate_rms(audio_frame) / level_cornerstone)

[0079] where calculate rms() is a function for calculating root mean square of an audio frame. In one illustrative example, the audio frame can be represented as audio_frame = 200ms * 16 and Level_comerstone = sqrt(l / 2).

[0080] FIG. 8 is a diagram 800 illustrating example input shapes and output shapes of the machine learning model architecture 200 of FIG. 2 and the encoder 400 of FIG. 4. For example,PATENTQualcomm Reference No. 24051 0WO22the diagram 800 includes a list of layers of an encoder 802 (e.g., the encoder 204 of FIG. 2 and the encoder 400 of FIG. 4) with corresponding input shapes and output shapes. The diagram 800 includes a recurrent machine learning network 804 of FIG. 8 with corresponding input shape and output shape. The input shapes and output shapes corresponding to the encoder 802 and the recurrent machine learning network 804 are provided as non-limiting examples. The encoder 802 and the recurrent machine learning network 804 can include different input shapes and output shapes than listed in the diagram 800.

[0081] FIG. 9 is a block diagram illustrating example training audio data 900 for training a machine learning model for wind noise detection and wind noise estimation. For example, the training audio data 900 can include one or more input audio segments 902. For example, the input audio segments can be time step sections of an audio signal (e.g., time step sections of 200 ms). The training audio data 900 can include a level label 904 associated with a strength of wind noise (e.g., in decibels) present in the audio segments. In some examples, the training audio data 900 can be augmented audio data modified to add wind noise to an audio signal.

[0082] In some examples, the training audio data 900, or the level label 904 can be provided to a machine learning model or engine for generating a one-hot level label 906 representation of the training audio data 900. For example, the machine learning model can generate the one-hot level label by transforming or processing the level label 904 into a vector. For example, generating the one-hot level label can include converting class indexes of training audio data into a binary representation where each class is represented by a unique binary vector. By way of example, each of the five one-hot level labels corresponds to a level label 904. The machine learning model can be trained using various loss training techniques, such as using cross-entropy loss functions, weighted kappa loss for ordinal data, etc.

[0083] In some aspects, training of one or more of the machine learning systems or neural networks described herein (e.g., such as the machine learning model architecture 200 of FIG. 2, the machine learning model 304 of FIG. 3, neural network 1100 of FIG. 11, the convolutional neural network 1200 of FIG. 12, among various other machine learning networks described herein) can be performed using online training (e.g., in some case on-device training), offline training, and / or various combinations of online and offline training. In some cases, online may refer to timePATENTQualcomm Reference No. 24051 0WO23periods during which the input data (e.g., such as the audio signal 202 of FIG. 2, etc.) is processed, for instance for performance of the wind noise detection and estimation implemented by the systems and techniques described herein. In some examples, offline may refer to idle time periods or time periods during which input data is not being processed. Additionally, offline may be based on one or more time conditions (e.g., after a particular amount of time has expired, such as a day, a week, a month, etc.) and / or may be based on various other conditions such as network and / or server availability, etc., among various others. In some aspects, offline training of a machine learning model (e.g., a neural network model) can be performed by a first device (e.g., a server device) to generate a pre-trained model, and a second device can receive the trained model from the second device. In some cases, the second device (e.g., a mobile device, an XR device, a vehicle or system / component of the vehicle, or other device) can perform online (or on-device) training of the pre-trained model to further adapt or tune the parameters of the model.

[0084] FIG. 10 is a flow chart illustrating an example of a process 1000 for detecting and estimating wind noise from an audio signal. The process 1000 can be performed by a computing device (e.g., SOC 100 of FIG. 1, computing device or computing system 1300 of FIG. 13, etc.) or by a component or system (e.g., the machine learning model architecture 200 of FIG. 2, the machine learning model 304 of FIG. 3, the encoder 204 of FIG. 2, the recurrent machine learning network 208 of FIG. 2, etc.) ,a chipset, one or more processors central processing units (CPUs), digital signal processors (DSPs), graphics processing units (GPUs), any other type of processor(s), any combination thereof, or other component or system) of the computing device. The operations of the process 1000 can be implemented as software components that are executed and run on one or more processors (e.g., processor 1310ofFIG. 13 or other processor(s)) of the computing device. Further, the transmission and reception of signals by the computing device in the process 1000 can be enabled, for example, by one or more antennas, one or more microphones, and / or one or more transceivers (e.g., wireless transceiver(s)).

[0085] At block 1002, the computing device (or component thereof) can process, using an encoder (e.g., the encoder 204 of FIG. 2, the encoder 400 of FIG. 4, etc.) of a machine learning model (e.g., the machine learning model architecture 200 of FIG. 2, the machine learning model 304 of FIG. 3, the neural network 1100 of FIG. 11, etc.), the audio signal to generate featuresPATENTQualcomm Reference No. 24051 0WO24associated with the audio signal. In another example, the encoder can include a one-dimensional convolutional neural network (e.g., the convolutional neural network 1200). For example, the encoder can receive the audio signal (or a tokenized representation or embedding representation of the audio signal) and determine or generate features associated with the audio signals. In such an example, the features can include properties or characteristics of the audio signals, such as features associated with the audio signal can include signal sub-band centroid (SSC) features of the audio signal. In further examples, the encoder can receive a time step section of the audio signal, and the features can be associated with the time step section of the audio signal. In some examples, the computing device can include one or more microphones configured to capture the audio signal.

[0086] In some aspects, the computing device can be triggered to detect wind signal based on movement. For example, the computing device can include a camera or a motion sensor to detect wind, such as detection movement of objects in an environment from wind. In such an example, the computing device can determine, based on movement detected by a sensor (e.g., one or more sensors), to process the audio signal. In some aspects, the machine learning model can be trained using training data generated based on wind noise added to a training audio signal. In further examples, the machine learning model can be trained using on-device training.

[0087] At block 1004, the computing device (or component thereof) can process, using a recurrent machine learning network of the machine learning model (e.g., , the recurrent machine learning network 208 of FIG. 2), the features associated with the audio signal to determine a classification of the audio signal based on the features. In some examples, the recurrent machine learning network can include a gated recurrent unit and a long short-term memory. In further examples, the classification of the audio signal can include a level from a plurality of levels. The levels can be associated with a probability the time step section of the audio signal includes the wind noise based on the features. For example, the levels can be the example levels of wind noise estimation of diagram 700 from FIG. 7. In further examples, the level can correspond to a range of wind noise values based on the features associated with the time step section of the audio signal.

[0088] At block 1006, the computing device (or component thereof) can determine, based on the classification of the audio signal, wind noise is present in the audio signal. In some examples, thePATENTQualcomm Reference No. 2405130WO25classification of the audio signal can include a level associated with a probability the audio signal includes the wind noise based on the features. In further examples, the computing device (or a component thereof) can determine the classification of the audio signal based on the features. In such an example, the classification can be associated with the level of the plurality of levels and each level of the plurality of levels can correspond to a range of wind noise values from a plurality of ranges. In a further example, the computing device (or component thereof) can determine, based on the classification of the audio signal and a comparison of the classification to a predetermined threshold level, the wind noise is present in the time step section of the audio signal.

[0089] As noted above, various aspects of the present disclosure can use machine learning models or systems. FIG. 11 is an illustrative example of a deep learning neural network 1100 that can be used to implement the machine learning based feature extraction and / or activity recognition (or classification) described above. An input layer 1120 includes input data. In one illustrative example, the input layer 1120 can include data representing the pixels of an input video frame. The neural network 1100 includes multiple hidden layers 1122a, 1122b, through 1122n. The hidden layers 1122a, 1122b, through 1122n include “n” number of hidden layers, where “n” is an integer greater than or equal to one. The number of hidden layers can be made to include as many layers as needed for the given application. The neural network 1100 further includes an output layer 1121 that provides an output resulting from the processing performed by the hidden layers 1122a, 1122b, through 1122n.

[0090] The neural network 1100 is a multi-layer neural network of interconnected nodes. Each node can represent a piece of information. Information associated with the nodes is shared among the different layers and each layer retains information as information is processed. In some cases, the neural network 1100 can include a feed-forward network, in which case there are no feedback connections where outputs of the network are fed back into itself. In some cases, the neural network 1100 can include a recurrent neural network, which can have loops that allow information to be carried across nodes while reading in input.

[0091] Information can be exchanged between nodes through node-to-node interconnections between the various layers. Nodes of the input layer 1120 can activate a set of nodes in the first hidden layer 1122a. For example, as shown, each of the input nodes of the input layer 1120 isPATENTQualcomm Reference No. 24051 0WO26connected to each of the nodes of the first hidden layer 1122a. The nodes of the first hidden layer 1122a can transform the information of each input node by applying activation functions to the input node information. The information derived from the transformation can then be passed to and can activate the nodes of the next hidden layer 1122b, which can perform their own designated functions. Example functions include convolutional, up-sampling, data transformation, and / or any other suitable functions. The output of the hidden layer 1122b can then activate nodes of the next hidden layer, and so on. The output of the last hidden layer 1122n can activate one or more nodes of the output layer 1121, at which an output is provided. In some cases, while nodes (e.g., node 1126) in the neural network 1100 are shown as having multiple output lines, a node has a single output and all lines shown as being output from a node represent the same output value.

[0092] In some cases, each node or interconnection between nodes can have a weight that is a set of parameters derived from the training of the neural network 1100. Once the neural network 1100 is trained, it can be referred to as a trained neural network, which can be used to classify one or more activities. For example, an interconnection between nodes can represent a piece of information learned about the interconnected nodes. The interconnection can have a tunable numeric weight that can be tuned (e.g., based on a training dataset), allowing the neural network 1100 to be adaptive to inputs and able to learn as more and more data is processed.

[0093] The neural network 1100 is pre-trained to process the features from the data in the input layer 1120 using the different hidden layers 1122a, 1122b, through 1122n in order to provide the output through the output layer 1121. In an example in which the neural network 1100 is used to reduce wind noise in an audio signal, the neural network 1100 can be trained using training data generated in the process described in FIG. 6.

[0094] In some cases, the neural network 1100 can adjust the weights of the nodes using a training process called backpropagation. As noted above, a b ackpropagation process can include a forward pass, a loss function, a backward pass, and a weight update. The forward pass, loss function, backward pass, and parameter update is performed for one training iteration. The process can be repeated for a certain number of iterations for each set of training images until the neural network 1100 is trained well enough so that the weights of the layers are accurately tuned.PATENTQualcomm Reference No. 24051 0WO27

[0095] For the example of reducing noise in audio signals, the forward pass can include passing training data through the neural network 1100. The weights are initially randomized before the neural network 1100 is trained. As an illustrative example, an audio signal can include an array of numbers representing a sequence of sounds. Each number in the array can include a numerical value representing sounds in sequence. In one example, the array is a one-dimensional sequence of numbers.

[0096] As noted above, for a first training iteration for the neural network 1100, the output will likely include values that do not give preference to any particular class due to the weights being randomly selected at initialization. For example, if the output is a vector with probabilities that the object includes different classes, the probability value for each of the different classes may be equal or at least very similar (e.g., for ten possible classes, each class may have a probability value of 0.1). With the initial weights, the neural network 1100 is unable to determine low level features and thus cannot make an accurate determination of what the classification of the object might be. A loss function can be used to analyze error in the output. Any suitable loss function definition can be used, such as the loss function described in the description of FIG. 5. Another example of a loss function includes the mean squared error (MSE), defined as Etotai= S “ (target — output)2. The loss can be set to be equal to the value of Etotai.

[0097] The loss (or error) will be high for the first training data since the actual values will be much different than the predicted output. The goal of training is to minimize the amount of loss so that the predicted output is the same as the training label. The neural network 1100 can perform a backward pass by determining which inputs (weights) most contributed to the loss of the network and can adjust the weights so that the loss decreases and is eventually minimized. A derivative of the loss with respect to the weights (denoted as dL / dW, where W are the weights at a particular layer) can be computed to determine the weights that contributed most to the loss of the network. After the derivative is computed, a weight update can be performed by updating all the weights of the filters. For example, the weights can be updated so that they change in the opposite directionof the gradient. The weight update can be denoted as w = where w denotes a weight,Wi denotes the initial weight, and q denotes a learning rate. The learning rate can be set to anyPATENTQualcomm Reference No. 24051 0WO28suitable value, with a high learning rate including larger weight updates and a lower value indicating smaller weight updates.

[0098] The neural network 1100 can include any suitable deep network. One example includes a convolutional neural network (CNN), which includes an input layer and an output layer, with multiple hidden layers between the input and out layers. The hidden layers of a CNN include a series of convolutional, nonlinear, pooling (for downsampling), and fully connected layers. The neural network 1100 can include any other deep network other than a CNN, such as an autoencoder, a deep belief nets (DBNs), a Recurrent Neural Networks (RNNs), among others.

[0099] FIG. 12 is an illustrative example of a convolutional neural network (CNN) 1200. FIG.12 provides an example for operation of a convolutional neural network (CNN) 1200 on images and video, however the structure of the convolutional neural network (CNN) may be further adapted to receive one-dimensional inputs such as audio signals. The input layer 1220 of the CNN 1200 includes data representing an image or frame. For example, the data can include an array of numbers representing the pixels of the image, with each number in the array including a value from 0 to 255 describing the pixel intensity at that position in the array. For example, the array can include a 28 x 28 x 3 array of numbers with 28 rows and 28 columns of pixels and 3 color components (e.g., red, green, and blue, or luma and two chroma components, or the like). The image can be passed through a convolutional hidden layer 1222a, an optional non-linear activation layer, a pooling hidden layer 1222b, and fully connected hidden layers 1222c to get an output at the output layer 1224. While only one of each hidden layer is shown in FIG. 12, one of ordinary skill will appreciate that multiple convolutional hidden layers, non-linear layers, pooling hidden layers, and / or fully connected layers can be included in the CNN 1200. The output can indicate a single class of an object or can include a probability of classes that best describe the object in the image.

[0100] The first layer of the CNN 1200 is the convolutional hidden layer 1222a. The convolutional hidden layer 1222a analyzes the image data of the input layer 1220. Each node of the convolutional hidden layer 1222a is connected to a region of nodes (pixels) of the input image called a receptive field. The convolutional hidden layer 1222a can be considered as one or more filters (each filter corresponding to a different activation or feature map), with each convolutionalPATENTQualcomm Reference No. 24051 0WO29iteration of a filter being a node or neuron of the convolutional hidden layer 1222a. For example, the region of the input image that a filter covers at each convolutional iteration would be the receptive field for the filter. In one illustrative example, if the input image includes a 28*28 array, and each filter (and corresponding receptive field) is a 5*5 array, then there will be 24*24 nodes in the convolutional hidden layer 1222a. Each connection between a node and a receptive field for that node learns a weight and, in some cases, an overall bias such that each node learns to analyze its particular local receptive field in the input image. Each node of the hidden layer 1222a will have the same weights and bias (called a shared weight and a shared bias). For example, the filter has an array of weights (numbers) and the same depth as the input. A filter will have a depth of 3 for the video frame example (according to three color components of the input image). An illustrative example size of the filter array is 5 x 5 x 3, corresponding to a size of the receptive field of a node.

[0101] The convolutional nature of the convolutional hidden layer 1222a is due to each node of the convolutional layer being applied to its corresponding receptive field. For example, a filter of the convolutional hidden layer 1222a can begin in the top-left corner of the input image array and can convolve around the input image. As noted above, each convolutional iteration of the filter can be considered a node or neuron of the convolutional hidden layer 1222a. At each convolutional iteration, the values of the filter are multiplied with a corresponding number of the original pixel values of the image (e.g., the 5x5 filter array is multiplied by a 5x5 array of input pixel values at the top-left corner of the input image array). The multiplications from each convolutional iteration can be summed together to obtain a total sum for that iteration or node. The process is next continued at a next location in the input image according to the receptive field of a next node in the convolutional hidden layer 1222a. For example, a filter can be moved by a step amount (referred to as a stride) to the next receptive field. The stride can be set to 1 or another suitable amount. For example, if the stride is set to 1, the filter will be moved to the right by 1 pixel at each convolutional iteration. Processing the filter at each unique location of the input volume produces a number representing the filter results for that location, resulting in a total sum value being determined for each node of the convolutional hidden layer 1222a.PATENTQualcomm Reference No. 24051 0WO30

[0102] The mapping from the input layer to the convolutional hidden layer 1222a is referred to as an activation map (or feature map). The activation map includes a value for each node representing the filter results at each location of the input volume. The activation map can include an array that includes the various total sum values resulting from each iteration of the filter on the input volume. For example, the activation map will include a 24 x 24 array if a 5 x 5 filter is applied to each pixel (a stride of 1) of a 28 x 28 input image. The convolutional hidden layer 1222a can include several activation maps in order to identify multiple features in an image. The example shown in FIG. 12 includes three activation maps. Using three activation maps, the convolutional hidden layer 1222a can detect three different kinds of features, with each feature being detectable across the entire image.

[0103] In some examples, a non-linear hidden layer can be applied after the convolutional hidden layer 1222a. The non-linear layer can be used to introduce non-linearity to a system that has been computing linear operations. One illustrative example of a non-linear layer is a rectified linear unit (ReLU) layer. A ReLU layer can apply the function f(x) = max (0, x) to all of the values in the input volume, which changes all the negative activations to 0. The ReLU can thus increase the non-linear properties of the CNN 1200 without affecting the receptive fields of the convolutional hidden layer 1222a.

[0104] The pooling hidden layer 1222b can be applied after the convolutional hidden layer 1222a (and after the non-linear hidden layer when used). The pooling hidden layer 1222b is used to simplify the information in the output from the convolutional hidden layer 1222a. For example, the pooling hidden layer 1222b can take each activation map output from the convolutional hidden layer 1222a and generates a condensed activation map (or feature map) using a pooling function. Max-pooling is one example of a function performed by a pooling hidden layer. Other forms of pooling functions be used by the pooling hidden layer 1222a, such as average pooling, L2-norm pooling, or other suitable pooling functions. A pooling function (e.g., a max-pooling filter, an L2-norm filter, or other suitable pooling filter) is applied to each activation map included in the convolutional hidden layer 1222a. In the example shown in FIG. 12, three pooling filters are used for the three activation maps in the convolutional hidden layer 1222a.PATENTQualcomm Reference No. 24051 0WO31

[0105] In some examples, max-pooling can be used by applying a max-pooling filter (e.g., having a size of 2x2) with a stride (e g., equal to a dimension of the filter, such as a stride of 2) to an activation map output from the convolutional hidden layer 1222a. The output from a maxpooling filter includes the maximum number in every sub-region that the filter convolves around. Using a 2x2 filter as an example, each unit in the pooling layer can summarize a region of 2x2 nodes in the previous layer (with each node being a value in the activation map). For example, four values (nodes) in an activation map will be analyzed by a 2x2 max-pooling filter at each iteration of the filter, with the maximum value from the four values being output as the “max” value. If such a max-pooling filter is applied to an activation filter from the convolutional hidden layer 1222a having a dimension of 24x24 nodes, the output from the pooling hidden layer 1222b will be an array of 12x12 nodes.

[0106] In some examples, an L2-norm pooling filter could also be used. The L2-norm pooling filter includes computing the square root of the sum of the squares of the values in the 2*2 region (or other suitable region) of an activation map (instead of computing the maximum values as is done in max-pooling) and using the computed values as an output.

[0107] Intuitively, the pooling function (e.g., max-pooling, L2-norm pooling, or other pooling function) determines whether a given feature is found anywhere in a region of the image and discards the exact positional information. This can be done without affecting results of the feature detection because, once a feature has been found, the exact location of the feature is not as important as its approximate location relative to other features. Max-pooling (as well as other pooling methods) offer the benefit that there are many fewer pooled features, thus reducing the number of parameters needed in later layers of the CNN 1200.

[0108] The final layer of connections in the network is a fully-connected layer that connects every node from the pooling hidden layer 1222b to every one of the output nodes in the output layer 1224. Using the example above, the input layer includes 28 x 28 nodes encoding the pixel intensities of the input image, the convolutional hidden layer 1222a includes 3*24*24 hidden feature nodes based on application of a 5*5 local receptive field (for the filters) to three activation maps, and the pooling hidden layer 1222b includes a layer of 3* 12* 12 hidden feature nodes based on application of max-pooling filter to 2*2 regions across each of the three feature maps.PATENTQualcomm Reference No. 24051 0WO32Extending this example, the output layer 1224 can include ten output nodes. In such an example, every node of the 3x12x12 pooling hidden layer 1222b is connected to every node of the output layer 1224.

[0109] The fully connected layer 1222c can obtain the output of the previous pooling hidden layer 1222b (which should represent the activation maps of high-level features) and determines the features that most correlate to a particular class. For example, the fully connected layer 1222c layer can determine the high-level features that most strongly correlate to a particular class and can include weights (nodes) for the high-level features. A product can be computed between the weights of the fully connected layer 1222c and the pooling hidden layer 1222b to obtain probabilities for the different classes. For example, if the CNN 1200 is being used to predict that an object in a video frame is a person, high values will be present in the activation maps that represent high-level features of people (e.g., two legs are present, a face is present at the top of the object, two eyes are present at the top left and top right of the face, a nose is present in the middle of the face, a mouth is present at the bottom of the face, and / or other features common for a person).

[0110] In some examples, the output from the output layer 1224 can include an M-dimensional vector (in the prior example, M=10). M indicates the number of classes that the CNN 1200 can choose from when classifying the object in the image. Other example outputs can also be provided. Each number in the M-dimensional vector can represent the probability the object is of a certain class. In one illustrative example, if a 10-dimensional output vector represents ten different classes of objects is [00 0.05 0.8 0 0.15 00 0 0], the vector indicates that there is a 5% probability that the image is the third class of object (e.g., a dog), an 80% probability that the image is the fourth class of object (e.g., a human), and a 15% probability that the image is the sixth class of object (e.g., a kangaroo). The probability for a class can be considered a confidence level that the object is part of that class.[oni] FIG. 13 is a diagram illustrating an example of a system for implementing certain aspects of the present technology. In particular, FIG. 13 illustrates an example of computing system 1300, which can be for example any computing device making up internal computing system, a remote computing system, a camera, or any component thereof in which the components of the system are in communication with each other using connection 1305. Connection 1305 can be a physicalPATENTQualcomm Reference No. 24051 0WO33connection using a bus, or a direct connection into processor 1310, such as in a chipset architecture. Connection 1305 can also be a virtual connection, networked connection, or logical connection.

[0112] In some aspects, computing system 1300 is a distributed system in which the functions described in this disclosure can be distributed within a datacenter, multiple data centers, a peer network, etc. In some aspects, one or more of the described system components represents many such components each performing some or all of the function for which the component is described. In some aspects, the components can be physical or virtual devices.

[0113] Example system 1300 includes at least one processing unit (CPU or processor) 1310 and connection 1305 that couples various system components including system memory 1315, such as read-only memory (ROM) 1320 and random access memory (RAM) 1325 to processor 1310. Computing system 1300 can include a cache 1313 of high-speed memory connected directly with, in close proximity to, or integrated as part of processor 1310.

[0114] Processor 1310 can include any general purpose processor and a hardware service or software service, such as services 1332, 1334, and 1336 stored in storage device 1330, configured to control processor 1310 as well as a special-purpose processor where software instructions are incorporated into the actual processor design. Processor 1310 may essentially be a completely self-contained computing system, containing multiple cores or processors, a bus, memory controller, cache, etc. A multi-core processor may be symmetric or asymmetric.

[0115] To enable user interaction, computing system 1300 includes an input device 1345, which can represent any number of input mechanisms, such as a microphone for speech, a touch-sensitive screen for gesture or graphical input, keyboard, mouse, motion input, speech, etc. Computing system 1300 can also include output device 1335, which can be one or more of a number of output mechanisms. In some instances, multimodal systems can enable a user to provide multiple types of input / output to communicate with computing system 1300. Computing system 1300 can include communications interface 1340, which can generally govern and manage the user input and system output. The communication interface may perform or facilitate receipt and / or transmission wired or wireless communications using wired and / or wireless transceivers, including those making use of an audio jack / plug, a microphone jack / plug, a universal serial bus (USB) port / plug, an Apple® Lightning® port / plug, an Ethernet port / plug, a fiber optic port / plug, a proprietary wired port / plug,PATENTQualcomm Reference No. 24051 0WO34a BLUETOOTH® wireless signal transfer, a BLUETOOTH® low energy (BLE) wireless signal transfer, an IBEACON® wireless signal transfer, a radio-frequency identification (RFID) wireless signal transfer, near-field communications (NFC) wireless signal transfer, dedicated short range communication (DSRC) wireless signal transfer, 802.11 Wi-Fi wireless signal transfer, wireless local area network (WLAN) signal transfer, Visible Light Communication (VLC), Worldwide Interoperability for Microwave Access (WiMAX), Infrared (IR) communication wireless signal transfer, Public Switched Telephone Network (PSTN) signal transfer, Integrated Services Digital Network (ISDN) signal transfer, 3G / 4G / 5G / LTE cellular data network wireless signal transfer, ad-hoc network signal transfer, radio wave signal transfer, microwave signal transfer, infrared signal transfer, visible light signal transfer, ultraviolet light signal transfer, wireless signal transfer along the electromagnetic spectrum, or some combination thereof. The communications interface 1340 may also include one or more Global Navigation Satellite System (GNSS) receivers or transceivers that are used to determine a location of the computing system 1300 based on receipt of one or more signals from one or more satellites associated with one or more GNSS systems. GNSS systems include, but are not limited to, the US-based Global Positioning System (GPS), the Russia-based Global Navigation Satellite System (GLONASS), the China-based BeiDou Navigation Satellite System (BDS), and the Europe-based Galileo GNSS. There is no restriction on operating on any particular hardware arrangement, and therefore the basic features here may easily be substituted for improved hardware or firmware arrangements as they are developed.

[0116] Storage device 1330 can be a non-volatile and / or non-transitory and / or computer-readable memory device and can be a hard disk or other types of computer readable media which can store data that are accessible by a computer, such as magnetic cassettes, flash memory cards, solid state memory devices, digital versatile disks, cartridges, a floppy disk, a flexible disk, a hard disk, magnetic tape, a magnetic strip / stripe, any other magnetic storage medium, flash memory, memristor memory, any other solid-state memory, a compact disc read only memory (CD-ROM) optical disc, a rewritable compact disc (CD) optical disc, digital video disk (DVD) optical disc, a blu-ray disc (BDD) optical disc, a holographic optical disk, another optical medium, a secure digital (SD) card, a micro secure digital (microSD) card, a Memory Stick® card, a smartcard chip, a EMV chip, a subscriber identity module (SIM) card, a mini / micro / nano / pico SIM card, another integrated circuit (IC) chip / card, random access memory (RAM), static RAM (SRAM), dynamicPATENTQualcomm Reference No. 24051 0WO35RAM (DRAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash EPROM (FLASHEPROM), cache memory (L1 / L2 / L3 / L4 / L5 / L#), resistive random-access memory (RRAM / ReRAM), phase change memory (PCM), spin transfer torque RAM (STT-RAM), another memory chip or cartridge, and / or a combination thereof.

[0117] The storage device 1330 can include software services, servers, services, etc., that when the code that defines such software is executed by the processor 1310, it causes the system to perform a function. In some aspects, a hardware service that performs a particular function can include the software component stored in a computer-readable medium in connection with the necessary hardware components, such as processor 1310, connection 1305, output device 1335, etc., to carry out the function.

[0118] As used herein, the term “computer-readable medium” includes, but is not limited to, portable or non-portable storage devices, optical storage devices, and various other mediums capable of storing, containing, or carrying instruction(s) and / or data. A computer-readable medium may include a non-transitory medium in which data can be stored and that does not include carrier waves and / or transitory electronic signals propagating wirelessly or over wired connections. Examples of a non-transitory medium may include, but are not limited to, a magnetic disk or tape, optical storage media such as compact disk (CD) or digital versatile disk (DVD), flash memory, memory or memory devices. A computer-readable medium may have stored thereon code and / or machine-executable instructions that may represent a procedure, a function, a subprogram, a program, a routine, a subroutine, an engine, a module, a software package, a class, or any combination of instructions, data structures, or program statements. A code segment may be coupled to another code segment or a hardware circuit by passing and / or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc. may be passed, forwarded, or transmitted using any suitable means including memory sharing, message passing, token passing, network transmission, or the like.

[0119] In some aspects the computer-readable storage devices, mediums, and memories can include a cable or wireless signal containing a bit stream and the like. However, when mentioned,PATENTQualcomm Reference No. 24051 0WO36non-transitory computer-readable storage media expressly exclude media such as energy, carrier signals, electromagnetic waves, and signals per se.

[0120] Specific details are provided in the description above to provide a thorough understanding of the aspects and examples provided herein. However, it will be understood by one of ordinary skill in the art that the aspects may be practiced without these specific details. For clarity of explanation, in some instances the present technology may be presented as including individual functional blocks including functional blocks comprising devices, device components, steps or routines in a method embodied in software, or combinations of hardware and software. Additional components may be used other than those shown in the figures and / or described herein. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form in order not to obscure the aspects in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the aspects.

[0121] Individual aspects may be described above as a process or method which is depicted as a flowchart, a flow diagram, a data flow diagram, a structure diagram, or a block diagram. Although a flowchart may describe the operations as a sequential process, many of the operations can be performed in parallel or concurrently. In addition, the order of the operations may be rearranged. A process is terminated when its operations are completed but could have additional steps not included in a figure. A process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc. When a process corresponds to a function, its termination can correspond to a return of the function to the calling function or the main function.

[0122] Processes and methods according to the above-described examples can be implemented using computer-executable instructions that are stored or otherwise available from computer-readable media. Such instructions can include, for example, instructions and data which cause or otherwise configure a general purpose computer, special purpose computer, or a processing device to perform a certain function or group of functions. Portions of computer resources used can be accessible over a network. The computer executable instructions may be, for example, binaries, intermediate format instructions such as assembly language, firmware, source code, etc. Examples of computer-readable media that may be used to store instructions, information used, and / orPATENTQualcomm Reference No. 24051 0WO37information created during methods according to described examples include magnetic or optical disks, flash memory, USB devices provided with non-volatile memory, networked storage devices, and so on.

[0123] Devices implementing processes and methods according to these disclosures can include hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof, and can take any of a variety of form factors. When implemented in software, firmware, middleware, or microcode, the program code or code segments to perform the necessary tasks (e.g., a computer-program product) may be stored in a computer-readable or machine-readable medium. A processor(s) may perform the necessary tasks. Typical examples of form factors include laptops, smart phones, mobile phones, tablet devices or other small form factor personal computers, personal digital assistants, rackmount devices, standalone devices, and so on. Functionality described herein also can be embodied in peripherals or add-in cards. Such functionality can also be implemented on a circuit board among different chips or different processes executing in a single device, by way of further example.

[0124] The instructions, media for conveying such instructions, computing resources for executing them, and other structures for supporting such computing resources are example means for providing the functions described in the disclosure.

[0125] In the foregoing description, aspects of the application are described with reference to specific aspects thereof, but those skilled in the art will recognize that the application is not limited thereto. Thus, while illustrative aspects of the application have been described in detail herein, it is to be understood that the inventive concepts may be otherwise variously embodied and employed, and that the appended claims are intended to be construed to include such variations, except as limited by the prior art. Various features and aspects of the above-described application may be used individually or jointly. Further, aspects can be utilized in any number of environments and applications beyond those described herein without departing from the broader spirit and scope of the specification. The specification and drawings are, accordingly, to be regarded as illustrative rather than restrictive. For the purposes of illustration, methods were described in a particular order. It should be appreciated that in alternate aspects, the methods may be performed in a different order than that described.PATENTQualcomm Reference No. 24051 0WO38

[0126] One of ordinary skill will appreciate that the less than (“<”) and greater than (“>”) symbols or terminology used herein can be replaced with less than or equal to (“<”) and greater than or equal to (“>”) symbols, respectively, without departing from the scope of this description.

[0127] Where components are described as being “configured to” perform certain operations, such configuration can be accomplished, for example, by designing electronic circuits or other hardware to perform the operation, by programming programmable electronic circuits (e.g., microprocessors, or other suitable electronic circuits) to perform the operation, or any combination thereof.

[0128] The phrase “coupled to” refers to any component that is physically connected to another component either directly or indirectly, and / or any component that is in communication with another component (e.g., connected to the other component over a wired or wireless connection, and / or other suitable communication interface) either directly or indirectly.

[0129] Claim language or other language reciting “at least one of’ a set and / or “one or more” of a set indicates that one member of the set or multiple members of the set (in any combination) satisfy the claim. For example, claim language reciting “at least one of A and B” or “at least one of A or B” means A, B, or A and B. In another example, claim language reciting “at least one of A, B, and C” or “at least one of A, B, or C” means A, B, C, or A and B, or A and C, or B and C, or A and B and C. The language “at least one of’ a set and / or “one or more” of a set does not limit the set to the items listed in the set. For example, claim language reciting “at least one of A and B” or “at least one of A or B” can mean A, B, or A and B, and can additionally include items not listed in the set of A and B.

[0130] The various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the aspects disclosed herein may be implemented as electronic hardware, computer software, firmware, or combinations thereof. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the describedPATENTQualcomm Reference No. 24051 0WO39functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

[0131] The techniques described herein may also be implemented in electronic hardware, computer software, firmware, or any combination thereof. Such techniques may be implemented in any of a variety of devices such as general purposes computers, wireless communication device handsets, or integrated circuit devices having multiple uses including application in wireless communication device handsets and other devices. Any features described as modules or components may be implemented together in an integrated logic device or separately as discrete but interoperable logic devices. If implemented in software, the techniques may be realized at least in part by a computer-readable data storage medium comprising program code including instructions that, when executed, performs one or more of the methods described above. The computer-readable data storage medium may form part of a computer program product, which may include packaging materials. The computer-readable medium may comprise memory or data storage media, such as random access memory (RAM) such as synchronous dynamic random access memory (SDRAM), read-only memory (ROM), non-volatile random access memory (NVRAM), electrically erasable programmable read-only memory (EEPROM), FLASH memory, magnetic or optical data storage media, and the like. The techniques additionally, or alternatively, may be realized at least in part by a computer-readable communication medium that carries or communicates program code in the form of instructions or data structures and that can be accessed, read, and / or executed by a computer, such as propagated signals or waves.

[0132] The program code may be executed by a processor, which may include one or more processors, such as one or more digital signal processors (DSPs), general purpose microprocessors, an application specific integrated circuits (ASICs), field programmable logic arrays (FPGAs), or other equivalent integrated or discrete logic circuitry. Such a processor may be configured to perform any of the techniques described in this disclosure. A general purpose processor may be a microprocessor; but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other suchPATENTQualcomm Reference No. 24051 0WO40configuration. Accordingly, the term “processor,” as used herein may refer to any of the foregoing structure, any combination of the foregoing structure, or any other structure or apparatus suitable for implementation of the techniques described herein.

[0133] Illustrative aspects of the present disclosure include:

[0134] Aspect 1 : An apparatus for wind noise detection, the apparatus comprising: one or more memories configured to store an audio signal; and one or more processors coupled to the one or more memories and configured to: process, using an encoder of a machine learning model, the audio signal to generate features associated with the audio signal; process, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; and determine, based on the classification of the audio signal, wind noise is present in the audio signal.

[0135] Aspect 2: The apparatus of Aspect 1, wherein the features are associated with a time step section of the audio signal.

[0136] Aspect 3: The apparatus of any of Aspects 1 to 2, wherein the classification of the audio signal includes a level from a plurality of levels, and wherein the level is associated with a probability the time step section of the audio signal includes the wind noise based on the features.

[0137] Aspect 4: The apparatus of any of Aspects 1 to 3, wherein the level corresponds to a range of wind noise values based on the features associated with the time step section of the audio signal.

[0138] Aspect 5: The apparatus of any of Aspects 1 to 4, wherein the one or more processors is configured to: determine the classification of the audio signal based on the features, wherein the classification is associated with the level of the plurality of levels, each level of the plurality of levels corresponding to a range of wind noise values from a plurality of ranges; and determine, based on the classification of the audio signal and a comparison of the classification to a predetermined threshold level, the wind noise is present in the time step section of the audio signal.PATENTQualcomm Reference No. 24051 0WO41

[0139] Aspect 6: The apparatus of any of Aspects 1 to 5, wherein the classification of the audio signal is a level associated with a probability the audio signal includes the wind noise based on the features.

[0140] Aspect 7: The apparatus of any of Aspects 1 to 6, wherein the one or more processors is configured to: determine, based on movement detected by a sensor, to process the audio signal.

[0141] Aspect 8: The apparatus of any of Aspects 1 to 7, further comprising one or more sensors configured to detect the movement.

[0142] Aspect 9: The apparatus of any of Aspects 1 to 8, wherein the encoder is a onedimensional convolutional neural network.

[0143] Aspect 10: The apparatus of any of Aspects 1 to 9, wherein the features include signal sub-band centroid (SSC) features of the audio signal.

[0144] Aspect 11: The apparatus of any of Aspects 1 to 10, wherein the recurrent machine learning network includes a gated recurrent unit (GRU) and a long short-term memory (LSTM).

[0145] Aspect 12: The apparatus of any of Aspects 1 to 11, wherein the machine learning model is trained using training data generated based on wind noise added to a training audio signal.

[0146] Aspect 13 : The apparatus of any of Aspects 1 to 12, wherein the machine learning model is trained using on-device training.

[0147] Aspect 14: The apparatus of any of Aspects 1 to 13, further comprising one or more microphones configured to capture the audio signal.

[0148] Aspect 15: A method for wind noise detection, the method comprising: processing, using an encoder of a machine learning model, the audio signal to generate features associated with the audio signal; processing, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; and determining, based on the classification of the audio signal, wind noise is present in the audio signal.PATENTQualcomm Reference No. 24051 0WO42

[0149] Aspect 16: The method of Aspect 15, wherein the features are associated with atime step section of the audio signal.

[0150] Aspect 17: The method of any of Aspects 15 to 16, wherein the classification of the audio signal includes a level from a plurality of levels, and wherein the level is associated with a probability the time step section of the audio signal includes the wind noise based on the features.

[0151] Aspect 18: The method of any of Aspects 15 to 17, wherein the level corresponds to a range of wind noise values based on the features associated with the time step section of the audio signal.

[0152] Aspect 19: The method of any of Aspects 15 to 18, further comprising: determining the classification of the audio signal based on the features, wherein the classification is associated with the level of the plurality of levels, each level of the plurality of levels corresponding to a range of wind noise values from a plurality of ranges; and determining, based on the classification of the audio signal and a comparison of the classification to a predetermined threshold level, the wind noise is present in the time step section of the audio signal.

[0153] Aspect 20: The method of any of Aspects 15 to 19, wherein the classification of the audio signal is a level associated with a probability the audio signal includes the wind noise based on the features.

[0154] Aspect 21: The method of any of Aspects 15 to 20, further comprising: determining, based on movement detected by a sensor, to process the audio signal.

[0155] Aspect 22: The method of any of Aspects 15 to 21, further comprising: detecting, using one or more sensors, the movement.

[0156] Aspect 23: The method of any of Aspects 15 to 22, wherein the encoder is a onedimensional convolutional neural network.

[0157] Aspect 24: The method of any of Aspects 15 to 23, wherein the features include signal sub-band centroid (SSC) features of the audio signal.PATENTQualcomm Reference No. 24051 0WO43

[0158] Aspect 25: The method of any of Aspects 15 to 24, wherein the recurrent machine learning network includes a gated recurrent unit (GRU) and a long short-term memory (LSTM).

[0159] Aspect 26: The method of any of Aspects 15 to 25, wherein the machine learning model is trained using training data generated based on wind noise added to a training audio signal.

[0160] Aspect 27: The method of any of Aspects 15 to 26, wherein the machine learning model is trained using on-device training.

[0161] Aspect 28: The method of any of Aspects 15 to 27, further comprising: capturing the audio signal using one or more microphones.

[0162] Aspect 29. A non-transitory computer-readable medium having stored thereon instructions that, when executed by at least one processor, cause the at least one processor to perform one or more of operations according to any of Aspects 15 to 28.

[0163] Aspect 30. An apparatus for wind noise detection , the apparatus comprising one or more means for performing operations according to any of Aspects 15 to 28.

Claims

PATENTQualcomm Reference No. 24051 0WO44CLAIMSWHAT IS CLAIMED IS:

1. An apparatus for wind noise detection, the apparatus comprising:one or more memories configured to store an audio signal; andone or more processors coupled to the one or more memories and configured to:process, using an encoder of a machine learning model, the audio signal to generate features associated with the audio signal;process, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; anddetermine, based on the classification of the audio signal, wind noise is present in the audio signal.

2. The apparatus of claim 1, wherein the features are associated with a time step section of the audio signal.

3. The apparatus of claim 2, wherein the classification of the audio signal includes a level from a plurality of levels, and wherein the level is associated with a probability the time step section of the audio signal includes the wind noise based on the features.

4. The apparatus of claim 3, wherein the level corresponds to a range of wind noise values based on the features associated with the time step section of the audio signal.

5. The apparatus of claim 3, wherein the one or more processors is configured to:determine the classification of the audio signal based on the features, wherein the classification is associated with the level of the plurality of levels, each level of the plurality of levels corresponding to a range of wind noise values from a plurality of ranges; andPATENTQualcomm Reference No. 24051 0WO45determine, based on the classification of the audio signal and a comparison of the classification to a predetermined threshold level, the wind noise is present in the time step section of the audio signal.

6. The apparatus of claim 1, wherein the classification of the audio signal is a level associated with a probability the audio signal includes the wind noise based on the features.

7. The apparatus of claim 1, wherein the one or more processors is configured to:determine, based on movement detected by a sensor, to process the audio signal.

8. The apparatus of claim 7, further comprising one or more sensors configured to detect the movement.

9. The apparatus of claim 1, wherein the encoder is a one-dimensional convolutional neural network.

10. The apparatus of claim 1, wherein the features include signal sub-band centroid (SSC) features of the audio signal.

11. The apparatus of claim 1, wherein the recurrent machine learning network includes a gated recurrent unit (GRU) and a long short-term memory (LSTM).

12. The apparatus of claim 1, wherein the machine learning model is trained using training data generated based on wind noise added to a training audio signal.

13. The apparatus of claim 12, wherein the machine learning model is trained using on-device training.

14. The apparatus of claim 1, further comprising one or more microphones configured to capture the audio signal.PATENTQualcomm Reference No. 24051 0WO4615. A method for wind noise detection, the method comprising:processing, using an encoder of a machine learning model, an audio signal to generate features associated with the audio signal;processing, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; anddetermining, based on the classification of the audio signal, wind noise is present in the audio signal.

16. The method of claim 15, wherein the features are associated with a time step section of the audio signal.

17. The method of claim 16, wherein the classification of the audio signal includes a level from a plurality of levels, and wherein the level is associated with a probability the time step section of the audio signal includes the wind noise based on the features.

18. The method of claim 17, wherein the level corresponds to a range of wind noise values based on the features associated with the time step section of the audio signal.

19. The method of claim 17, further comprising:determining the classification of the audio signal based on the features, wherein the classification is associated with the level of the plurality of levels, each level of the plurality of levels corresponding to a range of wind noise values from a plurality of ranges; and determining, based on the classification of the audio signal and a comparison of the classification to a predetermined threshold level, the wind noise is present in the time step section of the audio signal.

20. A non-transitory computer-readable medium having stored thereon instructions that, when executed by at least one processor, cause the at least one processor to:PATENTQualcomm Reference No. 24051 0WO47process, using an encoder of a machine learning model, an audio signal to generate features associated with the audio signal;process, using a recurrent machine learning network of the machine learning model, the features associated with the audio signal to determine a classification of the audio signal based on the features; anddetermine, based on the classification of the audio signal, wind noise is present in the audio signal.

Images