Individualized own voice detection in a hearing prosthesis

1. A method, comprising:

at one or more microphones of a hearing device, capturing input audio signals that include a voice of a recipient of the hearing device;

determining, from the input audio signals, a primary classification of a current sound environment associated with the input audio signals, wherein the primary classification indicates that the current sound environment includes speech signals;

after determining the primary classification, calculating, on the hearing device, time-varying features from the input audio signals; and

updating, based on an analysis of a plurality of the time-varying features, operation of an own voice detection decision tree of the hearing device.

2. The method of claim 1 , wherein the own voice detection decision tree is configured for classification of one or more time segments of input audio signals captured by the one or more microphones of the hearing device as either including the voice of the recipient or as including an external voice.

3. The method of claim 1 , wherein updating, based on an analysis of the plurality of the time-varying features, operation of an own voice detection decision tree, comprises:

obtaining a time-varying label that is time synchronized with the plurality of the time-varying features calculated on the hearing device; and

analyzing the plurality of the time-varying features and the time-varying label to generate updated decision tree weights for the own voice detection decision tree.

4. The method of claim 3 , wherein analyzing the plurality of the time-varying features and the time-varying label to generate updated decision tree weights for the own voice detection decision tree comprises:

executing a machine learning process to analyze the plurality of the time-varying features representative of the recipient's voice relative to values of the time-varying label at corresponding times.

5. The method of claim 3 , wherein the updated decision tree weights are generated at a computing device in communication with the hearing device, and wherein the method further comprises:

receiving the updated decision tree weights at the hearing device; and

instantiating the updated decision tree weights in the own voice detection decision tree of the hearing device.

6. The method of claim 5 , further comprising:

analyzing one or more input audio signals captured by the one or more microphones of the hearing device with the decision tree including the instantiated updated decision tree weights to classify time segments of the one or more input audio signals captured at the hearing device as either including the voice of the recipient or as including an external voice.

7. The method of claim 3 , wherein analyzing the plurality of the time-varying features and the time-varying label comprises:

analyzing the plurality of the time-varying features and the time-varying label on the hearing device; and

adjusting the decision tree weights based on the analysis of the plurality of the time-varying features and the time-varying label on the hearing device.

8. The method of claim 3 , wherein obtaining a time-varying label that is time synchronized with the plurality of the time-varying features comprises:

receiving a user input indicating which time segments of the input audio signals captured by the one or more microphones of the hearing device include the voice of the recipient.

9. The method of claim 8 , wherein receiving a user input comprises:

receiving an input from the recipient of the hearing device.

10. The method of claim 8 , wherein receiving a user input comprises:

receiving an input from an individual other than the recipient of the hearing device.

11. The method of claim 1 , wherein determining the primary classification of a current sound environment associated with the input audio signals, comprises:

determining the primary classification of the current sound environment based in part on an estimate of a harmonic signal power-to-total power ratio (STR) associated with the input audio signals.

12. The method of claim 1 , wherein determining the primary classification of a current sound environment associated with the input audio signals, comprises:

determining the primary classification of the current sound environment based in part on an estimate of a fundamental frequency (F 0 ) associated with the input audio signals.

13. A method, comprising:

receiving input audio signals at a device, wherein the input audio signals include speech of a recipient of the device;

calculating, on the device, time-varying features from the input audio signals;

analyzing a plurality of the time-varying features with an own voice detection decision tree on the device;

receiving label data associated the input audio signals, wherein the label data indicates which time segments of the input audio signals include the voice of the recipient;

analyzing the plurality of the time-varying features and the label data to generate updated weights for the own voice detection decision tree; and

updating the own voice detection decision tree with the updated weights.

14. The method of claim 13 , wherein the own voice detection decision tree is configured for classification of one or more time segments of input audio signals received at the device as either including the voice of the recipient or as including an external voice.

15. The method of claim 13 , wherein the label data is time-varying and time synchronized with the plurality of the time-varying features.

16. The method of claim 13 , wherein analyzing the plurality of the time-varying features and the label data to generate updated weights for the own voice detection decision tree comprises:

executing a machine learning process to generate the updated weights for the own voice detection decision tree based on the plurality of the time-varying features and the label data.

17. The method of claim 13 , wherein the updated decision tree weights are generated at a computing device in communication with the device, and wherein updating the own voice detection decision tree with the updated weights comprises:

receiving the updated decision tree weights at the device; and

instantiating the updated decision tree weights in the own voice detection decision tree of the device.

18. The method of claim 13 , further comprising:

analyzing one or more input audio signals received at the device with the own voice detection decision tree that has been updated with the updated weights to classify time segments of the input audio signals received at the device as either including the voice of the recipient or as including an external voice.

19. The method of claim 13 , wherein analyzing the plurality of the time-varying features and the label data to generate updated weights for the own voice detection decision tree comprises:

analyzing the plurality of the time-varying features and the label data on the device to generate the updated weights.

20. The method of claim 13 , wherein receiving label data associated the input audio signals comprises:

receiving a user input indicating which time segments of the input audio signals received at the device include the voice of the recipient.

21. The method of claim 20 , wherein receiving a user input comprises:

receiving an input from the recipient of the device.

22. The method of claim 20 , wherein receiving a user input comprises:

receiving an input from an individual other than the recipient of the device.

23. The method of claim 13 , wherein prior to analyzing a plurality of the time-varying features with an own voice detection decision tree on the device, the method comprises:

determining on the device, from the input audio signals, a primary classification of a current sound environment associated with the input audio signals, wherein the primary classification indicates that the current sound environment includes speech signals.

24. The method of claim 23 , wherein determining the primary classification of a current sound environment associated with the input audio signals, comprises:

determining the primary classification of the current sound environment based in part on an estimate of a harmonic signal power-to-total power ratio (STR) associated with the input audio signals.

25. The method of claim 23 , wherein determining the primary classification of a current sound environment associated with the input audio signals, comprises:

determining the primary classification of the current sound environment based in part on an estimate of a fundamental frequency (F 0 ) associated with the input audio signals.

26. A method, comprising:

receiving a plurality of time-varying features generated from input audio signals captured at one or more microphones of a device, wherein the input audio signals include a voice of a recipient of the device;

receiving label data associated the input audio signals, wherein the label data indicates which of a plurality of time segments of the input audio signals include the voice of the recipient;

analyzing the plurality of time-varying features and the label data to generate updated weights for an own voice detection decision tree on the device; and

updating the own voice detection decision tree with the updated weights to generate an updated an own voice detection decision tree.

27. The method of claim 26 , wherein the own voice detection decision tree is configured for classification of one or more time segments of input audio signals received at the device as either including the voice of the recipient or as including an external voice.

28. The method of claim 26 , wherein the label data is time-varying and time synchronized with the plurality of time-varying features.

29. The method of claim 26 , wherein analyzing the plurality of time-varying features and the label data to generate updated weights for an own voice detection decision tree on the device comprises:

executing a machine learning process to generate the updated weights for the own voice detection decision tree based on the plurality of time-varying features and the label data.

30. The method of claim 26 , wherein the updated decision tree weights are generated at a computing device in communication with the device, and wherein updating the own voice detection decision tree with the updated weights comprises:

sending the updated decision tree weights to the device.

31. The method of claim 26 , wherein analyzing the plurality of time-varying features and the label data to generate updated weights for an own voice detection decision tree on the device comprises:

analyzing the plurality of time-varying features and the label data on the device to generate the updated weights.

32. The method of claim 26 , wherein receiving label data associated the input audio signals comprises:

receiving a user input indicating which time segments of the input audio signals received at the device include the voice of the recipient.