VARIABLE-COMPONENT DEEP NEURAL NETWORK FOR ROBUST SPEECH RECOGNITION

• US 20160275947A1
• Filed: 09/09/2014
• Published: 09/22/2016
• Est. Priority Date: 09/09/2014
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: LI, Jinyu, GONG, Yifan, ZHAO, Rui
• Granted Patent: US 10,019,990 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G10L 15/16 using artificial neural networks
• 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 19/24 Variable rate codecs, e.g. for generating different qualities using a scalable representation such as hierarchical encoding or layered encoding
• CPC Class Codes: G10L 25/84 for discriminating voice from noise

Abstract

Systems and methods for speech recognition incorporating environmental variables are provided. The systems and methods capture speech to be recognized. The speech is then recognized utilizing a variable component deep neural network (DNN). The variable component DNN processes the captured speech by incorporating an environment variable. The environment variable may be any variable that is dependent on environmental conditions or the relation of the user, the client device, and the environment. For example, the environment variable may be based on noise of the environment and represented as a signal-to-noise ratio. The variable component DNN may incorporate the environment variable in different ways. For instance, the environment variable may be incorporated into weighting matrices and biases of the DNN, the outputs of the hidden layers of the DNN, or the activation functions of the nodes of the DNN.

Claims

1. A method for recognizing speech, the method comprising:

capturing speech input;

determining a value for an environment variable;

utilizing a deep neural network (DNN) to recognize the captured speech input, wherein one or more components of the DNN are modeled as a set of functions of the environment variable; and

producing an output of recognized speech.

2. The method of claim 1, wherein the one or more components of the DNN are at least one of the group consisting of one or more weighting matrices and one or more biases of the DNN.

3. The method of claim 1, wherein the one or more components of the DNN are one or more outputs of a hidden layer of the DNN.

4. The method of claim 1, wherein the one or more components of the DNN are one or more activation functions of one or more nodes in the DNN.

5. The method of claim 1, wherein the environment variable is based on a noise of an environment.

6. The method of claim 5, wherein the environment variable is a signal-to-noise ratio.

7. A system for recognizing speech, the system comprising:

a speech capture device;

a feature extraction module;

an environment variable module, wherein the environment variable module determines a value for an environment variable; and

a speech recognition decoder, wherein the speech recognition decoder utilizes a deep neural network (DNN) to recognize speech captured by the speech capture device, wherein one or more components of the DNN are modeled as a set of functions of the environment variable.

8. The system of claim 7, wherein the one or more components of the one or more components of the DNN are at least one of the group consisting of one or more weighting matrices and one or more biases of the DNN.

9. The system of claim 7, wherein the one or more components of the DNN are one or more outputs of a hidden layer of the DNN.

10. The system of claim 7, wherein the one or more components of the DNN are one or more activation functions of one or more nodes in the DNN.

11. The system of claim 7, wherein the environment variable is based on a noise of an environment.

12. The system of claim 11, wherein the environment variable is a signal-to-noise ratio.