System and methods for adapting neural network acoustic models

Patent metadata

Field	Value
Publication number	US-10366687-B2
Application number	US-201514965637-A
Country	US
Kind code	B2
Filing date	Dec 10, 2015
Priority date	Dec 10, 2015
Publication date	Jul 30, 2019
Grant date	Jul 30, 2019

Abstract

Techniques for adapting a trained neural network acoustic model, comprising using at least one computer hardware processor to perform: generating initial speaker information values for a speaker; generating first speech content values from first speech data corresponding to a first utterance spoken by the speaker; processing the first speech content values and the initial speaker information values using the trained neural network acoustic model; recognizing, using automatic speech recognition, the first utterance based, at least in part on results of the processing; generating updated speaker information values using the first speech data and at least one of the initial speaker information values and/or information used to generate the initial speaker information values; and recognizing, based at least in part on the updated speaker information values, a second utterance spoken by the speaker.

First claim

What is claimed is:

1. A method for adapting a trained neural network acoustic model, the method comprising:
using at least one computer hardware processor to perform:
generating speaker information values for a speaker;
generating speech content values from speech data corresponding to an utterance spoken by the speaker;
processing the speech content values and the speaker information values using the trained neural network acoustic model, the trained neural network acoustic model comprising a neural network and the processing comprising inputting the speaker information values to a partial layer of nodes of the neural network that is positioned in the neural network before a hidden layer of nodes of the neural network, the partial layer of nodes being configured to apply a transformation to the speaker information values based on parameters with which the partial layer of nodes are configured; and
generating updated parameters for the partial layer of nodes based on the processing.

2. The method of claim 1 , further comprising using the at least one computer hardware processor to perform:
generating second speech content values from second speech data corresponding to a second utterance spoken by the speaker; and
processing the second speech content values and the speaker information values using the trained neural network acoustic model and the updated parameters for the partial layer of nodes.

3. The method of claim 1 , wherein generating the speaker information values comprises generating an i-vector for the speaker.

4. The method of claim 3 , wherein generating the i-vector for the speaker comprises generating a normalized and quantized i-vector for the speaker.

5. The method of claim 1 , wherein the neural network comprises at least three layers.

6. At least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by at least one computer hardware processor, cause the at least one computer hardware processor to perform a method for adapting a trained neural network acoustic model, the method comprising:
generating speaker information values for a speaker based, at least in part, on statistics from speech data;
generating speech content values from speech data corresponding to an utterance spoken by the speaker;
processing the speech content values and the speaker information values using the trained neural network acoustic model, the trained neural network acoustic model comprising a neural network and the processing comprising inputting the speaker information values to a partial layer of nodes of the neural network that is positioned in the neural network before a hidden layer of nodes of the neural network, the partial layer of nodes being configured to apply a transformation to the speaker information values based on parameters with which the partial layer of nodes are configured; and
generating updated parameters for the partial layer of nodes based on the processing.

7. The at least one non-transitory computer-readable storage medium of claim 6 , where the method further comprises:
generating second speech content values from second speech data corresponding to a second utterance spoken by the speaker; and
processing the second speech content values and the speaker information values using the trained neural network acoustic model and the updated parameters for the partial layer of nodes.

8. The method of claim 6 , wherein generating the speaker information values comprises generating an i-vector for the speaker.

9. The method of claim 8 , wherein generating the i-vector for the speaker comprises generating a normalized and quantized i-vector for the speaker.

10. A system for adapting a trained neural network acoustic model, comprising:
at least one computer hardware processor; and
at least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by the at least one computer hardware processor, cause the at least one computer hardware processor to perform:
generating speaker information values for a speaker based, at least in part, on statistics from speech data;
generating speech content values from speech data corresponding to an utterance spoken by the speaker;
processing the speech content values and the speaker information values using the trained neural network acoustic model, the trained neural network acoustic model comprising a neural network and the processing comprising inputting the speaker information values to a partial layer of nodes of the neural network that is positioned in the neural network before a hidden layer of nodes of the neural network, the partial layer of nodes being configured to apply a transformation to the speaker information values based on parameters with which the partial layer of nodes are configured; and
generating updated parameters for the partial layer of nodes based on the processing.

11. The system of claim 10 , further comprising using the at least one computer hardware processor to perform:
generating second speech content values from second speech data corresponding to a second utterance spoken by the speaker; and
processing the second speech content values and the speaker information values using the trained neural network acoustic model and the updated parameters for the partial layer of nodes.

12. The system of claim 10 , wherein generating the speaker information values comprises generating an i-vector for the speaker.

13. The system of claim 12 , wherein generating the i-vector for the speaker comprises generating a normalized and quantized i-vector for the speaker.

14. The system of claim 10 , wherein the neural network comprises at least three layers.

Assignees

• Nuance Communications Inc

Inventors

• Zhan Puming
• Li Xinwei

Classifications

• G10L15/14

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

• G10L15/16 Primary

  using artificial neural networks · CPC title

• G10L17/02

  Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction · CPC title

• G10L15/075 Primary

  supervised, i.e. under machine guidance · CPC title

• G10L15/07

  to the speaker · CPC title