Method and system for recognizing speech commands using background and foreground acoustic models

Patent metadata

Field	Value
Publication number	US-9805715-B2
Application number	US-201314106634-A
Country	US
Kind code	B2
Filing date	Dec 13, 2013
Priority date	Jan 30, 2013
Publication date	Oct 31, 2017
Grant date	Oct 31, 2017

Abstract

A method of recognizing speech commands includes generating a background acoustic model for a sound using a first sound sample, the background acoustic model characterized by a first precision metric. A foreground acoustic model is generated for the sound using a second sound sample, the foreground acoustic model characterized by a second precision metric. A third sound sample is received and decoded by assigning a weight to the third sound sample corresponding to a probability that the sound sample originated in a foreground using the foreground acoustic model and the background acoustic model. The method further includes determining if the weight meets predefined criteria for assigning the third sound sample to the foreground and, when the weight meets the predefined criteria, interpreting the third sound sample as a portion of a speech command. Otherwise, recognition of the third sound sample as a portion of a speech command is forgone.

First claim

What is claimed is:

1. A method of recognizing a speech command, comprising:
at an electronic device with one or more processors and memory:
generating a background acoustic model for a sound using a first sound sample, wherein the sound in the background acoustic model is represented as monophones and the background acoustic model is characterized by a first precision metric that is less than a first predefined inverse variance value, and wherein the first precision metric corresponds to an inverse of a variance of the background acoustic model;
generating a foreground acoustic model for the sound using a second sound sample, wherein the sound in the foreground acoustic model is represented as triphones and the second sound sample characterized by higher precision phonemes than the first sound sample, resulting in the foreground acoustic model being characterized by a second precision metric that is greater than a second predefined inverse variance value, and wherein the second precision metric corresponds to an inverse of a variance of the foreground acoustic model;
receiving a third sound sample corresponding to the sound;
decoding the third sound sample by:
using the foreground acoustic model and the background acoustic model, assigning a first weight to the third sound sample corresponding to a probability that the sound sample originated in a foreground;
determining if the first weight meets predefined criteria for assigning the third sound sample to the foreground;
in accordance with a determination that the first weight meets the predefined criteria, interpreting the third sound sample as a portion of a speech command originating from a user of the electronic device; and
in accordance with a determination that the first weight does not meet the predefined criteria, rejecting the third sound sample as having originated in the background rather than having originated from the user.

2. The method of claim 1 , wherein:
the foreground acoustic model for the sound includes a hidden Markov model; and
the background acoustic model for the sound includes a hidden Markov model.

3. The method of claim 1 , wherein the foreground acoustic model for the sound and the background acoustic model for the sound comprise a Gaussian mixture model.

4. The method of claim 1 , wherein the first predefined inverse variance value and the second predefined inverse variance value are equal.

5. The method of claim 1 , wherein the first predefined inverse variance value is less than the second predefined inverse variance value.

6. The method of claim 1 , further including:
constructing a decoding network having a first path corresponding to the foreground acoustic model and a second path corresponding to the background acoustic model;
assigning a second weight to the third sound sample corresponding to probability that the sound sample originated in a background;
wherein the first weight corresponds to a transfer probability of the third sound sample through the first path and the second weight corresponds to a transfer probability of the third sound sample through the second path.

7. The method of claim 6 , wherein the predefined criteria include determining whether the first weight is greater than the second weight.

8. The method of claim 1 , wherein the predefined criteria include determining whether the first weight corresponds to a probability greater than 0.5.

9. An electronic device, comprising:
one or more processors;
memory; and
one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including an operating system and instructions that when executed by the one or more processors cause the electronic device to:
generate a background acoustic model for a sound using a first sound sample, wherein the sound in the background acoustic model is represented as monophones and the background acoustic model is characterized by a first precision metric that is less than a first predefined inverse variance value, and wherein the first precision metric corresponds to an inverse of a variance of the background acoustic model;
generate a foreground acoustic model for the sound using a second sound sample, wherein the sound in the foreground acoustic model is represented as triphones and the second sound sample characterized by higher precision phonemes than the first sound sample, resulting in the foreground acoustic model being characterized by a second precision metric that is greater than a second predefined inverse variance value, and wherein the second precision metric corresponds to an inverse of a variance of the foreground acoustic model;
receive a third sound sample corresponding to the sound;
decode the third sound sample by:
using the foreground acoustic model and the background acoustic model, assigning a first weight to the third sound sample corresponding to a probability that the sound sample originated in a foreground;
determining if the first weight meets predefined criteria for assigning the third sound sample to the foreground;
in accordance with a determination that the first weight meets the predefined criteria, interpreting the third sound sample as a portion of a speech command originating from a user of the electronic device; and
in accordance with a determination that the first weight does not meet the predefined criteria, rejecting the third sound sample as having originated in the background rather than having originated from the user.

10. The electronic device of claim 9 , wherein:
the foreground acoustic model for the sound includes a hidden Markov model; and
the background acoustic model for the sound includes a hidden Markov model.

11. The electronic device of claim 9 , wherein the foreground acoustic model for the sound and the background acoustic model for the sound comprise a Gaussian mixture model.

12. The electronic device of claim 9 , wherein the first predefined inverse variance value and the second predefined inverse variance value are equal.

13. The electronic device of claim 9 , wherein the first predefined inverse variance value is less than the second predefined inverse variance value.

14. The electronic device of claim 9 , wherein the instructions further cause the device to:
construct a decoding network having a first path corresponding to the foreground acoustic model and a second path corresponding to the background acoustic model;
assign a second weight to the third sound sample corresponding to probability that the sound sample originated in a background;
wherein the first weight corresponds to a transfer probability of the third sound sample through the first path and the second weight corresponds to a transfer probability of the third sound sample through the second path.

15. The electronic device of claim 14 , wherein the predefined criteria include determining whether the first weight is greater than the second weight.

16. The electronic device of claim 9 , wherein the predefined criteria include determining whether the first weight corresponds to a probability greater than 0.5.

17. A non-transitory computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by an electronic device with one or more processors and memory, cause the electronic device to:
generate a background acoustic model for a sound using a first sound sample, wherein the sound in the background acoustic model is represented as monophones and the background acoustic m

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Yue Shuai
• Lu Li
• Zhang Xiang
• Xie dadong
• Liu Haibo
• Chen Bo
• Liu Jian

Classifications

• G10L15/083

  Recognition networks (G10L15/142, G10L15/16 take precedence) · CPC title

• G10L15/14 Primary

  using statistical models, e.g. Hidden Markov Models [HMMs] (G10L15/18 takes precedence) · CPC title

• G10L15/32

  Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems · CPC title

• G10L2015/088

  Word spotting · CPC title

• G10L15/063

  Training · CPC title