Audio output masking for improved automatic speech recognition

Patent metadata

Field	Value
Publication number	US-9704478-B1
Application number	US-201314094591-A
Country	US
Kind code	B1
Filing date	Dec 2, 2013
Priority date	Dec 2, 2013
Publication date	Jul 11, 2017
Grant date	Jul 11, 2017

Abstract

Features are disclosed for filtering portions of an output audio signal in order to improve automatic speech recognition on an input signal which may include a representation of the output signal. A signal that includes audio content can be received, and a frequency or band of frequencies can be selected to be filtered from the signal. The frequency band may correspond to a desired frequency band for speech recognition. An input signal can be obtained comprising audio data corresponding to a user utterance and presentation of the output signal. Automatic speech recognition can be performed on the input signal. In some cases, an acoustic model trained for use with such frequency band filtering may be used to perform speech recognition.

First claim

What is claimed is:

1. A system comprising:
one or more computing devices configured to at least:
receive an audio signal;
determine a frequency band based at least partly on a likelihood that audio input data regarding a user utterance will be present within the frequency band of a subsequent input signal;
generate a filtered output signal by filtering a portion of audio output data within the frequency band from the audio signal, wherein the filtered output signal is generated prior to receiving an input signal comprising audio data regarding the user utterance and presentation of the filtered output signal, and wherein filtering the portion of audio output data from the audio signal reduces energy of the audio signal in the frequency band;
generate audio using the filtered output signal;
receive the input signal, wherein the input signal comprises audio data regarding both the user utterance and presentation of the filtered output signal;
select an acoustic model of a plurality of acoustic models based at least partly on the acoustic model being associated with the frequency band; and
perform speech recognition using the input signal and the acoustic model to generate speech recognition results.

2. The system of claim 1 , wherein the audio signal comprises music, and wherein a genre of music is used to determine the frequency band.

3. The system of claim 1 , wherein the portion of the audio output data within the frequency band is filtered from the audio signal using a weighted filter.

4. The system of claim 1 , wherein the one or more computing devices are further configured to determine a plurality of different frequency bands, wherein audio output data in each of the plurality of different frequency bands is to be filtered from the audio signal.

5. The device of claim 1 , wherein the one or more computing devices are further configured to determine the frequency band to filter based at least partly on information regarding acoustic echo cancellation.

6. The system of claim 1 , wherein the acoustic model is selected based on the acoustic model being trained to recognize subword units using audio data within the frequency band.

7. A computer-implemented method comprising:
as implemented by a computing device comprising one or more processors configured to execute specific instructions,
receiving a first signal comprising data regarding audio content;
determining a frequency band within which audio data regarding a user utterance is expected to be present in an input signal;
generating an output signal comprising a portion of the first signal, wherein the output signal is generated prior to receiving the input signal, wherein the input signal comprises audio data corresponding to the user utterance and presentation of the output signal, and wherein the output signal excludes a portion of the first signal having a frequency within the frequency band;
receiving the input signal, wherein the input signal comprises audio data corresponding to the user utterance and presentation of the output signal, and wherein a portion of the input signal comprising audio data corresponding to the user utterance has a frequency within the frequency band; and
providing the input signal to a speech recognizer.

8. The computer-implemented method of claim 7 , wherein the portion of the first signal having the frequency within the frequency band is excluded from the output signal using a weighted filter.

9. The computer-implemented method of claim 7 , further comprising determining the frequency band based at least partly on a characteristic of the first signal.

10. The computer-implemented method of claim 9 , wherein the characteristic is determined using one of metadata associated with the first signal or an analysis of at least a portion of the first signal.

11. The computer-implemented method of claim 7 , further comprising determining the frequency band based at least partly on a vocal characteristic of a user.

12. The computer-implemented method of claim 11 , wherein the vocal characteristic is associated with one of: gender, age, language, dialect, or accent.

13. The computer-implemented method of claim 7 , further comprising determining the frequency band based at least partly on a desired signal-to-noise ratio of the output signal for the frequency band.

14. The computer-implemented method of claim 7 , wherein generating the output signal comprises excluding a second portion of the first signal, wherein the second portion is excluded based on a frequency of the second portion being within a second frequency band, wherein the second frequency band is different than the first frequency band, and wherein the second portion is different than the portion.

15. The computer-implemented method of claim 6 , further comprising selecting an acoustic model of a plurality of acoustic models based at least partly on the frequency band.

16. The computer-implemented method of claim 15 , wherein the acoustic model was trained using training data, and wherein at least a portion of audio data in the frequency band of the training data was filtered from the training data.

17. A device comprising:
means for receiving a first signal comprising data regarding audio content;
means for determining a frequency band within which audio data regarding a user utterance is expected to be present in an input signal;
means for generating an output signal comprising a portion of the first signal, wherein the output signal is generated prior to receiving the input signal, wherein the input signal comprises audio data corresponding to the user utterance and presentation of the output signal, and wherein the output signal excludes a portion of the first signal having a frequency within the frequency band;
means for receiving the input signal, wherein the input signal comprises audio data corresponding to the user utterance and presentation of the output signal, and wherein a portion of the input signal comprising audio data corresponding to the user utterance has a frequency within the frequency band; and
means for providing the input signal to a speech recognizer.

18. The device of claim 17 , wherein the portion of the first signal having the frequency within the frequency band is filtered using a weighted filter.

19. The device of claim 17 , further comprising means for determining the frequency band based at least partly on a characteristic of the first signal.

20. The device of claim 19 , wherein the characteristic is determined using one of metadata associated with the first signal or an analysis of at least a portion of the first signal.

21. The device of claim 17 , wherein the means for generating the output signal generates the output signal by excluding a second portion of the first signal, wherein the second portion is excluded based on a frequency of the second portion being within a second frequency band, wherein the second frequency band is different than the first frequency band, and wherein the second portion is different than the portion.

22. The device of claim 17 , wherein the portion of the first signal having the frequency in the frequency band is filtered in anticipation of receiving a user utterance.

23. The device of claim 17 , further comprising means for selecting an acoustic model of a plurality of acoustic models based at least partly on the frequency band.

24. The device of claim 23 , wherein the acoustic model is trained to prioritize the f

Assignees

• Amazon Tech Inc

Inventors

• Vitaladevuni Shiv Naga Prasad
• Chhetri Amit Singh
• Hilmes Phillip Ryan
• Prasad Rohit

Classifications

• G10L15/02 Primary

  Feature extraction for speech recognition; Selection of recognition unit · CPC title

• G10L21/0232 Primary

  Processing in the frequency domain · CPC title

• G10L2021/02082

  the noise being echo, reverberation of the speech · CPC title

• G10L15/00

  Speech recognition (G10L17/00 takes precedence) · CPC title