Transmission of noise parameters for improving automatic speech recognition

• US 8,983,844 B1
• Filed: 07/31/2012
• Issued: 03/17/2015
• Est. Priority Date: 07/31/2012
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Thomas, Ryan P., Strom, Nikko
• Primary Examiner(s): Vo, Huyen X.
• Application Number:
  US13/562,476
• Time in Patent Office: 959 Days
• Field of Search: 704/275, 704/226, 704/256, 704/256.2, 704/233, 704/235, 704/231, 704/270, 704/270.1, 379/406.01
• US Class Current: 704/270.1
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: G10L 21/0208 Noise filtering

Abstract

Methods and systems for transmission of noise parameters for improving automatic speech recognition are disclosed. A system includes one or more microphones, wherein each microphone is configured to produce an audio signal. The system also includes a noise reduction module configured to generate a noise-reduced audio signal and a noise parameter. Furthermore, the system includes a transmitter configured to transmit, to a computing device, the noise-reduced audio signal and a noise parameter. The computing device may use the noise parameter in obtaining a model to use for performing automatic speech recognition.

Claims

1. A method comprising:

receiving, by a server device comprising a processor, a noise-reduced audio signal and a noise parameter from a user device, wherein the noise parameter provides information relating to how noise was reduced in the noise-reduced audio signal;

selecting, by the server device, a first automatic speech recognition (ASR) model from a plurality of ASR models wherein;

each ASR model of the plurality of ASR models is associated with a respective noise parameter model, and

the first ASR model is selected based at least in part on a comparison between the noise parameter and the noise parameter model associated with the first ASR model; and

performing, by the server device, ASR processing on the noise-reduced audio signal using the first ASR model to generate ASR results.

2. The method of claim 1, wherein the noise parameter is a vector.

3. The method of claim 1, wherein the first ASR model is an acoustic model.

4. The method of claim 1, wherein the comparison is one of a Euclidean distance, a likelihood ratio, or a Mahalanobis distance.

5. The method of claim 1, wherein a first noise parameter model associated with the first ASR model is a Gaussian mixture model.

6. The method of claim 1, wherein the first ASR model minimizes the distance between the noise parameter and the first noise parameter model associated with each ASR model of the plurality of ASR models.

7. The method of claim 1, wherein the noise parameter comprises one or more of a signal to noise ratio, a list of amplitudes and delays, a list of weights, or a list of filter coefficients.

8. The method of claim 1, further comprising transmitting the ASR results to the user device.

9. A system comprising:

one or more microphones configured to produce an audio signal;

a processor to execute a noise reduction module configured to generate a noise-reduced audio signal and a noise parameter based on the audio signal produced by the one or more microphones, wherein the noise parameter provides information relating to how noise was reduced in the noise-reduced audio signal; and

a transmitter configured to transmit, to a server device, the noise-reduced audio signal and the noise parameter, wherein the noise parameter is used by the server device to select an automatic speech recognition (ASR) model to perform ASR processing on the noise-reduced audio signal.

10. The system of claim 9, wherein the server device is further configured to:

compute a plurality of feature vectors from the noise-reduced audio signal;

append the noise parameter to at least one feature vector of the plurality of feature vectors; and

perform the ASR processing using the obtained ASR model and the plurality of feature vectors, wherein the obtained ASR model was trained using at least one feature vector that comprised a noise parameter computed from training data.

11. The system of claim 9, wherein the noise-reduced audio signal is generated by applying a noise reduction algorithm to the audio signals, and wherein the noise-reduction algorithm comprise one of echo cancellation, dereverberation, beamforming, blind source separation, noise cancellation, or spectral shaping.

12. The system of claim 9, wherein the server device is configured to:

obtain an automatic speech recognition (ASR) model using the noise parameter; and

perform ASR processing using the obtained ASR model and the noise-reduced signal to generate ASR results.

13. The system of claim 12, wherein the server device is further configured to obtain an ASR model by selecting an ASR model from a plurality of ASR models.

14. The system of claim 12, wherein the server device is further configured to obtain an ASR model by adapting an existing ASR model.

15. The system of claim 9, wherein the computing device is a server device configured to:

apply one or more noise reduction algorithms to the noise-reduced audio signal to generate a second noise-reduced audio signal and a second noise parameter;

obtain an automatic speech recognition (ASR) model using the second noise parameter; and

perform ASR processing using the obtained ASR model and the second noise-reduced signal to generate ASR results.

16. The system of claim 9, wherein the ASR model is one of a neural network-based acoustic model or a Gaussian mixture model.

17. A non-transitory computer-readable storage medium having instructions that, when executed by a processing device, cause the processing device to perform operations comprising:

receiving, by a server device, a noise-reduced audio signal and a noise parameter from a user device, wherein the noise parameter provides information relating to how noise was reduced in the noise-reduced audio signal;

obtaining an automatic speech recognition (ASR) model using the noise parameter;

performing ASR processing on the noise-reduced audio signal using the obtained ASR model; and

transmitting a result of the ASR processing to the user device.

18. The non-transitory computer-readable storage medium of claim 17, wherein obtaining the ASR model using the noise parameter comprises selecting the ASR model from a plurality of ASR models using the noise parameter.

19. The non-transitory computer-readable storage medium of claim 17, wherein obtaining the ASR model using the noise parameter comprises modifying an existing ASR model using the noise parameter.

20. The non-transitory computer-readable storage medium of claim 17, wherein the noise-reduced audio signal is generated by applying a noise reduction algorithm to an audio signal produced by one or more microphones of the user device, and wherein the noise reduction algorithm is one of echo cancellation, dereverberation, beamforming, blind source separation, noise cancellation, or spectral shaping.

21. The non-transitory computer-readable storage medium of claim 17, wherein the noise parameter comprises one or more of a signal to noise ratio, a list of amplitudes and delays, a list of weights, or a list of filter coefficients.