Machine learning based self-utterance removal

Abstract

The present invention relates to a process that includes receiving an audio signal including a user's speech component and a noise component, filtering the audio signal with an ego-speech filter that utilizes eigen-user vectors to filter the speech component, the eigen-user vectors being determined based on the user's speech input during an offline training session and representing eigen-condensed representations of the user's speech patterns, and outputting a filtered audio signal in which the user's speech component has been substantially removed from the audio signal. The eigen-user vectors are obtained by machine learning using speech input from multiple speakers.

Description

[Technical field]

[0001] Claiming priority This application claims priority to U.S. Patent Application No. 17 / 032,801, filed September 25, 2020, which is incorporated by reference in its entirety.

[0002] The present disclosure relates generally to wearable hearing assistive devices and, more particularly, to a machine learning based approach for removing a user's speech signal in a wearable hearing assistive device. [Background technology]

[0003] Wearable hearing assistive devices can significantly improve a user's hearing experience. For example, such devices typically employ one or more microphones and amplification components to amplify a desired sound(s), such as the voice of another person speaking to the user. In addition, such devices may employ techniques such as active noise reduction (ANR) to combat undesirable environmental noise. Although wearable hearing assistive devices can be in a variety of form factors, e.g., headphones, earbuds, audio glasses, etc., filtering out unwanted acoustic signals, such as the user's own voice signal, continues to present a variety of technical challenges. Summary of the Invention

[0004] All embodiments and features mentioned below can be combined in any technically possible manner.

[0005] Systems and approaches are disclosed that employ a wearable hearing assistive device to remove a user's speech. Some implementations include receiving an audio signal, where the audio signal includes a user's speech component and a noise component, filtering the audio signal with an own-speech filter that utilizes an eigenuser vector to filter the speech component, where the eigenuser vector is determined based on a user's voice input, and outputting a filtered audio signal in which the user's speech component is substantially removed from the audio signal.

[0006] In an additional particular implementation, a system is provided that includes a memory and a processor coupled to the memory and configured to remove user speech for a hearing assistive device according to a method that includes receiving an audio signal, the audio signal including a user speech component and a noise component; filtering the audio signal with an own-speech filter that utilizes an eigenuser vector to filter the speech component, the eigenuser vector being determined based on a user voice input; and outputting a filtered audio signal in which the user speech component is substantially removed from the audio signal.

[0007] Implementations may include one or any combination of the following features.

[0008] In some cases, the eigenuser vector is determined during an offline session and comprises a unique condensed representation of the user's speech patterns.

[0009] In other cases, the unique compressed representation includes one of a d-vector representation or an i-vector representation.

[0010] In some aspects, the self-speech filter includes a machine learning model that is trained offline according to a method that includes providing a set of unique user vectors, each corresponding to a unique user, and an associated mixed audio signal for each unique user, where each associated mixed audio signal includes a speech component and a corresponding noise component for the unique user; inputting the mixed audio signal associated with the selected unique user vector to the machine learning model; training the machine learning model to remove the speech component from the associated mixed audio signal based on the selected unique user vector; and repeating the inputting and training for each unique user vector.

[0011] In certain implementations, the method includes post-processing the filtered audio signal to generate an enhanced filtered audio signal, and outputting the enhanced filtered audio signal to an electro-acoustic transducer to generate an acoustic signal.

[0012] In some cases, the post-processing includes at least one of speech enhancement, signal-to-noise ratio (SNR) improvement, beamforming, or active noise reduction.

[0013] In certain aspects, the hearing assistance device includes a head worn device having at least one earphone and an accessory device in communication with the head worn device.

[0014] In some implementations, the self-speech filter is included within the accessory device.

[0015] In certain cases, the self-speech filter is included in a head-worn device.

[0016] In some implementations, an offline session is used to determine eigenuser vectors using a speech characterization system that includes a machine learning model trained using a database of speech inputs from multiple speakers.

[0017] In some aspects, the self-talk filter includes at least one audio processing enhancement selected from the group consisting of signal-to-noise ratio (SNR) improvement, beamforming, or active noise reduction.

[0018] In another aspect, a system is provided that performs a method including receiving an audio signal, the audio signal including a target speech component of a non-device user and a noise component; filtering the audio signal with a target speech enhancer that utilizes an eigenuser vector to substantially pass the target speech component, the eigenuser vector being determined based on a voice input of the non-device user; and outputting a speech-enhanced audio signal including substantially only the target speech component.

[0019] Two or more of the features described in this disclosure, including the features described in this Summary section, may be combined to form implementations not specifically described herein.

[0020] The details of one or more implementations are set forth in the accompanying drawings and the description below. Other features, objects, and advantages will become apparent from the description and drawings, and from the claims. [Brief description of the drawings]

[0021] [Figure 1] FIG. 1 is a block diagram of a wearable hearing assistance device, according to various implementations. [Diagram 2] 1 illustrates a machine learning platform for implementing self-utterance removal, according to various implementations. [Diagram 3] 1 illustrates an alternative machine learning platform for performing self-utterance removal, according to various implementations. [Figure 4] 1 illustrates a machine learning platform for performing targeted speech enhancement, according to various implementations. [Diagram 5]1 illustrates an example of a wearable hearing assistive device, according to various implementations.

[0022] It should be noted that the drawings of the various implementations are not necessarily to scale. The drawings are intended to depict only typical aspects of the disclosure, and therefore should not be considered as limiting the scope of the implementations. In the drawings, like numbering represents like elements between the drawings. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] Various implementations describe solutions for removing a user's own voice ("self-speech removal") in a wearable hearing assistive device. Typically, when using a hearing assistive device, a user may be annoyed or otherwise irritated by the amplification of the user's own voice. However, amplification of others' voices is important for audibility.

[0024] In hearing assistive devices, such as hearing aids, audio augmented reality systems, and systems that utilize a remote microphone streaming (e.g., from a phone or other device) to headphones, sound is sent to the ear via two different paths. The first path is the "direct path," where the sound travels around the device or headphones and directly into the ear canal. In the second, "processing path," the audio travels through the hearing assistive device or headphones, is processed, and then delivered through a driver (i.e., an electrostatic transducer or speaker) to the ear canal.

[0025] A significant factor in the performance of any hearing assistive device when processing real-time audio for human consumption is the latency introduced by the algorithms that process the signal along the processing path. Latency is defined as the delay (usually measured in milliseconds) between when the audio enters the device and when it leaves. Excessive latency in the system can significantly degrade the perceived quality of the algorithms and the user experience.

[0026] If the overall system latency exceeds a certain threshold (e.g., on the order of 5-20 ms), it can be very confusing and annoying to the user. More precisely, this threshold is determined by the human sensitivity to the latency of their own voice in the processing path. However, if the user's own voice is not in the processing path, the user can tolerate the latency much better, and the overall system latency threshold increases significantly (e.g., to around 80-150 ms).

[0027] This disclosure describes implementations for using machine learning based processing to remove a user's own voice (i.e., self-speech) from the processing path, thereby effectively increasing the latency budget for audio processing.

[0028] Although generally described with respect to hearing assistance devices, the solutions disclosed herein are intended to be applicable to a wide variety of wearable audio devices, i.e., devices that are at least partially worn by a user near at least one of the user's ears and structured to provide amplified audio to at least one of the ears. Other such implementations may include headphones, two-way communication headsets, earphones, earbuds, hearing aids, audio glasses, wireless headsets (also known as "earsets"), and ear protectors. The presentation of specific implementations is intended to facilitate understanding through the use of examples and should not be construed as limiting either the scope of the disclosure or the applicability of the claims.

[0029] Furthermore, the solutions disclosed herein are applicable to wearable audio devices that provide two-way audio communication, one-way audio communication (i.e., acoustic output of audio electronically provided by another device), or no communication. Furthermore, what is disclosed herein is applicable to wearable audio devices that are connected to other devices wirelessly, connected to other devices via electrically and / or optically conductive cables, or not connected to any other devices. These teachings are applicable to wearable audio devices having a physical structure configured to be worn near either one or both ears of a user, including, but not limited to, headphones with one or two earpieces, over-the-head headphones, behind-the-neck headphones, headsets with a communication microphone (e.g., a boom microphone), in-ear or behind-ear hearing aids, wireless headsets (i.e., earsets), audio glasses, a single earphone or a pair of earphones, as well as hats, helmets, clothing, or any physical structure incorporating one or two earpieces to enable audio communication and / or ear protection.

[0030] In an example implementation, the processed audio may include any natural or artificial sound (or acoustic signal), and the microphone may include one or more microphones capable of capturing sound and converting it into an electronic signal.

[0031] In various implementations, the wearable audio devices (e.g., hearing assistance devices) described herein may incorporate active noise reduction (ANR) functionality, which may include one or both of feedback-based ANR and feedforward-based ANR, in addition to possibly providing audio that has been processed via typical hearing aid signal processing such as pass-through audio and dynamic range compression.

[0032] In addition, the solutions disclosed herein are intended to be applicable to a wide variety of accessory devices, i.e., devices that can communicate with a wearable audio device and assist in processing audio signals. Exemplary accessory devices include smartphones, Internet of Things (IoT) devices, computing devices, dedicated electronics, vehicles, computerized agents, carrying cases, charging cases, smart watches, other wearable devices, etc.

[0033] In various implementations, the wearable audio device (e.g., a hearing assistance device) and the accessory device communicate wirelessly, for example, using Bluetooth® or other wireless protocols, and in some implementations, the wearable audio device and the accessory device are within a few meters of each other.

[0034] 1 illustrates an example implementation of a wearable hearing assistive device 100 that utilizes a machine learning (ML) based approach to output a processed audio signal 126 in which self-speech 118 has been removed or substantially removed. As shown, the device 100 includes a set of microphones 114 configured to receive a mixed acoustic input 115 that includes a mixture of self-speech 118 and noise 120. The noise 120 generally includes all other acoustic input other than the self-speech 118, such as other speech, background sounds, environmental sounds, music, etc. The microphone input 116 receives the mixed audio signal from the microphones 114 and passes the mixed audio signal 128 to the audio processing system 102.

[0035] The audio processing system 102 includes a self-speech filter 104 that includes a trained machine learning (ML) model configured to process the mixed audio signal 128. More specifically, the self-speech filter 104 removes the user's self-speech 118 using, for example, an eigen-user vector 112 created during an enrollment process (or stage) 122 based on the user's voice. In some implementations, the eigen-user vector 112 that includes a eigen-compressed representation of the user's speech pattern is determined during the enrollment process 122. In other implementations, the eigen-user vector 112 can be calculated by the device 100. The ML process for generating the eigen-user vector 112 and the self-speech filter 104 is further described herein, for example, with reference to FIG. 2. Once the mixed audio signal 128 has been processed with the self-speech filter 104 to remove the user's (i.e., the user of the wearable hearing assistive device 100) self-speech, the post-processing algorithms 106 can further enhance the filtered signal by providing, for example, speech enhancement, signal-to-noise (SNR) improvement, beamforming, active noise reduction, etc. The resulting signal is then amplified by the amplifier system 108 and output via the electro-acoustic transducer 124.

[0036] 2 shows an example overview of a machine learning based process utilized to implement the self-utterance filter 104 and to calculate the eigenuser vector 112, such as that shown with the device 100 (FIG. 1). The process may be implemented in three stages, including an offline training stage, an enrollment stage, and an inference stage. An implementation of the inference stage is described in FIG. 1, where the device 100 filters the user's self-utterance during operation of the device 100 by the user. The enrollment stage can be performed as a one-time or on-demand operation, for example, when a new user starts using the device 100. In various implementations, the offline training stage is responsible for training two ML-based models required for the other stages, namely the speech characterization system 103 and the self-utterance filter 104.

[0037] In a particular implementation, the speech characterization system 103 (first model) is trained using the user utterance database (DB) 202 and the associated eigen-user vector DB 204, for example, using a neural network or the like. For example, the first model can be trained using user utterance / expected eigen-user vector pairs. The user utterances are input to the speech characterization system 103 to generate eigen-user vectors. The generated eigen-user vectors can be compared with expected eigen-user vectors from the eigen-user vector DB 204, and the results are fed back to adjust the first model. Training can continue until each user utterance from the user utterance database (DB) 202 input to the model reliably outputs its expected eigen-user vector. Once trained, the resulting speech characterization system 103 can be used in an enrollment phase to map the utterances of a new (or enrolling) user to an eigen-user vector 112 that is explicitly associated with the new user. The eigen-user vector 112 generally consists of a unique compressed representation of the user's speech pattern. The eigen-compressed representation may include, for example, a d-vector representation or an i-vector representation. As will be readily appreciated in the art, various algorithms exist for extracting d-vectors and i-vectors based on a user's voice samples.

[0038] The self-speech filter 104 (second model), also trained offline using a neural network or the like, is trained to remove the person's speech from the mixed audio signal (i.e., an audio signal containing both self-speech 118 and noise 120) using the person's eigenuser vector. For example, the model can be trained by inputting a mixed audio signal from the mixed audio signal DB 206 with the associated eigenuser vector to obtain a self-speech filtered signal, which can be compared to an expected signal in the self-speech filtered signal database 208. The results can be fed back to adjust and tune the second model. Once trained, the resulting self-speech filter 104 can be deployed (e.g., on the device 100) in an inference stage.

[0039] During the enrollment phase, the user is instructed to record a short audio clip of himself speaking, for example, a set of predefined utterances. This can be done on the device 100 itself, or on an accessory device such as a mobile phone or co-processor. A system or guidance may be utilized to ensure that the recording contains only the user's voice and not other interfering audio or utterances. For example, the user may be instructed to perform the utterance in a quiet space, and / or the resulting audio clip can be analyzed to ensure that no undesirable noise is present. The recorded enrollment utterances are then passed through a speech characterization system 103 (either on the device 100 itself, on a companion device, in the cloud, etc.) and the unique user vector 112 of the enrolled user is extracted.

[0040] Once the eigen-user vector 112 is extracted, it can be loaded and stored, for example, on the device 100 to perform the inference stage. During this stage, the mixed audio including the user's own speech 212 is first sent to the self-speech filter 104 along with the user's eigenvector 112 to remove the user's own voice from the processing path. The resulting self-speech filtered audio 214 is then passed to any downstream post-processing algorithms 106 such as speech enhancement, SNR improvement, beamforming, ANR, etc. Finally, the self-speech filtered enhanced audio 218 is sent to the driver to be played in the user's ear canal.

[0041] Due to the fact that the user's own voice is removed from the processing path during the inference stage, the user can tolerate much higher processing latency than with an audio signal that includes self-speech. This higher tolerance to latency enables more complex downstream post-processing algorithms 106, for example algorithms implemented on accessories such as smartphones or wireless co-processor devices, that accumulate latency not only from the algorithmic processing but also from the over-the-air transmission of the audio.

[0042] Figure 3 shows an alternative embodiment in which the self-talk filter 104 and post-processing algorithms 106 of Figure 2 are combined into a single system, namely the self-talk filter and post-processing system 105. In this way, processing improvements such as speech enhancement, SNR improvement, beamforming, ANR, etc. are addressed by a single system or machine learning training model.

[0043] 4 illustrates a further implementation in which a similar machine learning platform is utilized to provide speech-enhanced audio 215, either stand-alone or in addition to the self-speech filtering platform described herein. In this case, the platform creates eigen-user vectors designed to pass speech from one or more non-device users (i.e., target speech). Instead of training a model to remove speech corresponding to the eigen-user vectors (i.e., the self-speech filter 104 of FIG. 2), the model, in this case the target speech enhancer 107, is trained to retain the target speech of the non-device users. The target speech enhancer 107 is trained by processing a mixed audio signal (including target speech and noise) with the associated eigen-user vectors to reliably generate speech-enhanced audio 209 that includes only the target speech. The speech characterization system 103 is trained in the same manner as the self-speech filtering platform.

[0044] Thus, a user of the device 101 can choose to only hear speech from a contact by registering the contact in the speech characterization system 103 and loading the contact's unique user vector 112 into the target speech enhancer 107. When mixed audio 213 including the contact's speech, i.e., the target speech, is presented to the device 101, the target speech enhancer 107 generates speech-enhanced audio 215 including substantially only the contact's speech. Processed enhanced audio 219 can then be generated from the post-processing algorithm 106. In an alternative embodiment, the post-processing algorithm 106 can be incorporated into the target speech enhancer 107.

[0045] The process can be extended to accommodate multiple targets through which speech is allowed to pass by enrolling multiple users, each of which generates a unique user vector 112. The multiple unique user vectors 112 can then be input to the target speech enhancer 107.

[0046] It should be understood that the devices 100, 101 (FIGS. 1-4) shown and described according to various implementations may be structured to be worn by a user to provide audio output proximate at least one of the user's ears. The devices 100, 101 may have any of several form factors, such as configurations incorporating a single earpiece to provide audio to only one of the user's ears, other configurations incorporating a pair of earpieces to provide audio to both of the user's ears, and other configurations incorporating one or more standalone speakers to provide audio to the user's surrounding environment. An exemplary wearable audio device is shown and described in further detail in U.S. Patent No. 10,194,259 ("Directional Audio Selection"), filed February 28, 2018, which is incorporated herein by reference in its entirety.

[0047] In an example implementation, the acoustic input 115 (FIG. 1) may include any ambient acoustic signal, including, for example, an acoustic signal generated by a user of the wearable hearing assistive device, including natural or artificial sounds (or acoustic signals). The microphone 114 may include one or more microphones (e.g., one or more microphone arrays including a feedforward microphone and / or a feedback microphone) capable of capturing and converting sound into an electronic signal.

[0048] Although several examples have been provided herein for removing a user's voice in a hearing assistive device, it should be understood that other approaches, or combinations of the described approaches, may be used.

[0049] FIG. 5 is a schematic diagram of an exemplary wearable hearing assistive device 300 (in one example form factor) including a self-speech removal system housed within a housing 302. It should be understood that the exemplary wearable hearing assistive device 300 can include some or all of the components and functions described with respect to the devices 100, 101 shown and described with reference to FIGS. 1-4. The self-speech removal system and / or the target speech enhancer can be one of, or can run on, one or more of the various electronics 304 within the housing 302. In certain embodiments, some or all of the self-speech or enhancement algorithms may be implemented in an accessory 330 configured to communicate with the wearable hearing assistive device 300. In this example, the wearable audio device 300 is an audio headset that includes two earphones (e.g., in-ear headphones also referred to as "earbuds") 312, 314. The earphones 312, 314 are tethered to a housing 302 configured to rest on a user's neck (e.g., a neckband), although other configurations, including wireless configurations, may be utilized. Each illustrated earphone 312, 314 includes a body 316 that may include a case formed of one or more plastic or composite materials. The body 316 may include a nozzle 318 for insertion into the entrance of the user's ear canal and a support member 320 for holding the nozzle 318 in a resting position in the user's in-ear. In addition to the self-speech removal system, the control unit 302 may include other electronics 304, such as an amplifier, a battery, user controls, a voice activity detection (VAD) device, and the like.

[0050] In some implementations, as described above, the separate accessory 330 can include a communication system 332, for example, for wireless communication with the device 300, and remote processing 334 for providing some or all of the functionality described herein, such as the self-speech filter 104, the registration process 122, the post-processing algorithms 106, etc. The accessory 330 can be implemented in many embodiments. In one embodiment, the accessory 330 includes a standalone device. In another embodiment, the accessory 330 includes a user-supplied smartphone that utilizes a software application that enables remote processing 334 while using the smartphone hardware for the communication system 332. In another embodiment, the accessory 330 can be implemented in a charging case of the device 300. In another embodiment, the accessory 330 can be implemented in a companion microphone accessory that also performs other functions such as off-head beamforming and wireless streaming of beamformed audio to the device 300. Other wearable device forms, such as around-ear headphones, audio glasses, open-ear audio devices, etc., can also be implemented.

[0051] With reference to FIG. 1, the set of microphones 114 may include an in-ear microphone. With reference to FIG. 5, such an in-ear microphone may be integrated into the earphone body 316, for example, into the nozzle 318. The in-ear microphone may also be used to perform feedback active noise reduction (ANR) and voice pickup for communication, which may be performed in other electronic devices 304. The user's own voice in the ear canal is substantially different from that received from a microphone placed outside the ear canal due to bone and tissue conduction of the own voice. With reference to FIG. 2, the in-ear microphone, together with the other microphones in the set 114, may provide unique characteristics to the mixed audio including the mixed audio signal DB 206, the user enrolled utterances 210, and the user's own speech 212. At the same time, the in-ear and out-ear microphones provide more information to the speech characterization system 103 and improve the performance of the self-speech filter 104.

[0052] According to various implementations, a hearing assistive device is provided that filters a user's own speech to improve performance. In particular, the own speech filter 104, which utilizes the eigenuser vectors 112, removes the user's voice from the mixed audio signal.

[0053] It will be understood that one or more of the functions of the described systems may be implemented as hardware and / or software, and that the various components may include communication paths connecting the components by any conventional means (e.g., wired and / or wireless connections). For example, one or more non-volatile devices (e.g., centralized or distributed devices such as flash memory devices) may store and / or execute the programs, algorithms, and / or parameters of one or more of the described systems of devices. Also, the functionality or portions thereof, and various modifications thereof (hereinafter "functionality") described herein may be implemented at least in part via a computer program product (e.g., a computer program tangibly embodied in an information carrier, such as one or more non-transitory machine-readable media, for execution by or for controlling the operation of one or more data processing devices (e.g., programmable processors, computers, multiple computers, and / or programmable logic components, etc.).

[0054] The computer program may be written in any form of programming language, including compiled or interpreted languages, and it may be arranged in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment. The computer program may be arranged to be executed on one computer or on multiple computers at one site, or may be distributed across multiple sites and interconnected by a network.

[0055] Actions associated with carrying out all or a portion of the functions may be performed by one or more programmable processors executing one or more computer programs to carry out the functions. All or a portion of the functions may be implemented as special purpose logic circuitry, such as field programmable gate arrays (FPGAs) and / or application-specific integrated circuits (ASICs). Processors suitable for executing computer programs include, by way of example, both general purpose and special purpose microprocessors, as well as any one or more processors of any kind of digital computer. In general, a processor may receive instructions and data from a read-only memory, a random access memory, or both. The components of a computer include a processor for executing instructions and one or more memory devices for storing instructions and data.

[0056] It should be noted that while the implementations described herein utilize a microphone system to collect input signals, it is understood that any type of sensor, e.g., accelerometer, thermometer, optical sensor, camera, etc., may be utilized separately or in addition to the microphone system to collect input signals.

[0057] Additionally, actions associated with implementing all or a portion of the functionality described herein may be performed by one or more networked computing devices that may be connected via a network, e.g., one or more wired and / or wireless networks, such as a local area network (LAN), a wide area network (WAN), a personal area network (PAN), an Internet-connected device and / or network, and / or cloud-based computing (e.g., cloud-based servers).

[0058] In various implementations, electronic components described as "coupled" may be linked via conventional wired and / or wireless means such that the electronic components can communicate data with one another. Furthermore, subcomponents within a given component may be considered to be linked via conventional paths, although not necessarily illustrated.

[0059] Although multiple implementations have been described, it is understood that additional modifications may be made without departing from the scope of the inventive concepts described herein, and thus, other implementations are within the scope of the following claims. [Explanation of symbols]

[0060] 100 Wearable hearing aid devices 102 Audio Processing System 103 Speech Characterization System 104 Self-talk filter 105 Aftertreatment System 106 Post-processing Algorithms 107 Target Speech Enhancer 108 Amplifier System 112 Eigenuser Vectors 112 Eigenvectors 114 Microphone 115 Mixed Acoustic Input 116 Microphone Input 118 Self-talk 120 Noise 122 Registration Process 124 Electroacoustic Transducer 126 Audio Signals 128 mixed audio signals 202 User Speech Database (DB) 208 Signal Database 209 Speech Enhancement Audio 210 User Registration Speech 212 Self-talk 213 Mixed Audio 214 Audio 215 Speech Enhancement Audio 218 Processed Enhanced Audio 300 Wearable hearing aid devices 302 Control Unit 304 Electronic equipment 312, 314 Earphones 316 Main Body 318 Nozzle 320 Support member 330 Accessories 332 Communication Systems 334 Remote Processing

Claims

1. 1. A method for removing user speech for a hearing assistive device, comprising: receiving an audio signal, the audio signal including a user speech component and a noise component; filtering the audio signal with an ego-speech filter that utilizes an eigenuser vector to filter the speech component, the eigenuser vector being determined during an offline session and comprising an ego-compressed representation of the user's speech pattern; and outputting a filtered audio signal in which the speech component of the user is substantially removed from the audio signal.

2. The method of claim 1 , wherein the eigen-compressed representation comprises one of a d-vector representation or an i-vector representation.

3. The self-utterance filter comprises a machine learning model that is trained offline according to a method, the method comprising: providing a set of unique user vectors, each corresponding to a unique user, and an associated mixed audio signal for each unique user, each of the associated mixed audio signals including a speech component of the unique user and a corresponding noise component; inputting the selected eigenuser vectors and the associated mixed audio signal into the machine learning model; training the machine learning model to remove the speech component from the associated mixed audio signal based on the selected eigenuser vectors; and repeating the inputting and training for each eigen-user vector.

4. post-processing the filtered audio signal to generate an enhanced filtered audio signal; outputting the enhanced filtered audio signal to an electro-acoustic transducer to generate an acoustic signal; The method of claim 1 further comprising:

5. The method of claim 4 , wherein the post-processing includes at least one of speech enhancement, signal-to-noise ratio (SNR) improvement, beamforming, or active noise reduction.

6. The method of claim 1 , wherein the hearing assistance device comprises a head worn device having at least one earphone and an accessory device in communication with the head worn device.

7. The method of claim 6 , wherein the self-speech filter is included in the accessory device.

8. The method of claim 6 , wherein the self-speech filter is included in the head-worn device.

9. The method of claim 6 , wherein the hearing assistive device includes an earphone having an in-ear microphone that communicates with an accessory device.

10. 2. The method of claim 1 , wherein the offline session used to determine the eigenuser vectors utilizes a speech characterization system comprising a machine learning model trained using a database of speech inputs from multiple speakers.

11. 1. A system comprising: Memory, a processor coupled to the memory and configured to remove user utterances for the hearing assistive device according to a method, the method comprising: receiving an audio signal, the audio signal including a user speech component and a noise component; filtering the audio signal with an ego-speech filter that utilizes an eigenuser vector to filter the speech component, the eigenuser vector being determined during an offline session and comprising an ego-compressed representation of the user's speech pattern; and outputting a filtered audio signal in which the speech component of the user is substantially removed from the audio signal.

12. The system of claim 11 , wherein the eigen-compressed representation comprises one of a d-vector representation or an i-vector representation.

13. The self-utterance filter comprises a machine learning model that is trained offline according to a method, the method comprising: a set of eigenuser vectors, each corresponding to a unique user; providing an associated mixed audio signal for each of the unique users, each associated mixed audio signal including a speech component and a corresponding noise component for the unique user; inputting the selected eigenuser vectors and the associated mixed audio signal into the machine learning model; training the machine learning model to remove the speech component from the associated mixed audio signal based on the selected eigenuser vectors; and repeating the inputting and training for each eigen-user vector.

14. post-processing the filtered audio signal to generate an enhanced filtered audio signal; outputting the enhanced filtered audio signal to an electro-acoustic transducer to generate an acoustic signal; The system of claim 11 further comprising:

15. The system of claim 14 , wherein the post-processing comprises at least one of speech enhancement, signal-to-noise ratio (SNR) improvement, beamforming, or active noise reduction.

16. The system of claim 11 , wherein the hearing assistance device comprises a head-worn device having at least one earphone and an accessory device in communication with the head-worn device.

17. The system of claim 16 , wherein the self-speech filter is included in one of the accessory device or the head-worn device.

18. The system of claim 16 , wherein the earphone includes an in-ear microphone that communicates with the accessory device.

19. 12. The system of claim 11, wherein the offline session used to determine the eigenuser vectors utilizes a speech characterization system comprising a machine learning model trained using a database of speech inputs from multiple speakers.

20. 12. The system of claim 11, wherein the self-talk filter includes at least one audio processing enhancement selected from the group consisting of signal-to-noise ratio (SNR) improvement, beamforming, or active noise reduction.

Images