SPEECH RECOGNITION METHOD AND SYSTEM BASED ON USER PERSONALIZED INFORMATION

• US 20150348542A1
• Filed: 12/20/2013
• Published: 12/03/2015
• Est. Priority Date: 12/28/2012
• Status: Active Grant

• Current Assignee: Iflytek Co., Ltd.
• Original Assignee: Iflytek Co., Ltd.
• Inventors: PAN, Qinghua, HE, Tingting, HU, Guoping, HU, Yu, LIU, Qingfeng
• Granted Patent: US 9,564,127 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 3/167 Audio in a user interface, e.g. using voice commands for navigating, audio feedback
• CPC Class Codes: G10L 15/083 Recognition networks G10L15/142, G10L15/16 take precedence
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: G10L 19/00 Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis in musical instruments G10H
• CPC Class Codes: G10L 2015/227 of the speaker; Human-factor methodology

Abstract

The present invention relates to a speech recognition method and system based on user personalized information. The method comprises the following steps: receiving a speech signal; decoding the speech signal according to a basic static decoding network to obtain a decoding path on each active node in the basic static decoding network, wherein the basic static decoding network is a decoding network associated with a basic name language model; if a decoding path enters a name node in the basic static decoding network, network extending is carried out on the name node according to an affiliated static decoding network of a user, wherein the affiliated static decoding network is a decoding network associated with a name language model of a particular user; and returning a recognition result after the decoding is completed. The recognition accuracy rate of user personalized information in continuous speech recognition may be raised by using the present invention.

Claims

1. A speech recognition method based on user personalized information, characterized in that comprising:

receiving a speech signal;

decoding the speech signal according to a basic static decoding network to obtain a decoding path on each active node in the basic static decoding network, wherein the basic static decoding network is a decoding network formed by extending the words in a basic name language model into corresponding acoustic units, the basic name language model is a language model for describing the statistical probability between common words and the statistical probability between common words and names;

if a decoding path enters a name node in the basic static decoding network, extending an extra network on the name node according to an affiliated static decoding network of a user, wherein the affiliated static decoding network is a decoding network formed by extending the words in a name language model of a particular user into corresponding acoustic units, the name language model of a particular user is a language model for describing the statistical probability of the specific name associated with the user; and

returning a recognition result after the decoding is completed.

2. The method according to claim 1, wherein the method further comprises the following steps:

determining the affiliated static decoding network of the user before the speech signal is decoded according to the basic static decoding network;

or

determining the affiliated static decoding network of the user after a decoding path enters a name node in the basic static decoding network.

3. The method according to claim 2, wherein the step of determining the affiliated static decoding network of the user comprises:

determining the user identity according to a feature of the speech signal, and then determining the affiliated static decoding network of the user according to the user identity;

or

determining the user identity according to an equipment code or account number of the user, and then determining the affiliated static decoding network of the user according to the user identity.

4. The method according to claim 1, wherein the method further comprises the following steps:

generating a basic name language model and a name language model of a particular user; and

respectively building a basic static decoding network associated with the basic name language model and an affiliated static decoding network associated with the name language model of the particular user.

5. The method according to claim 4, wherein the step of generating a basic name language model comprises:

respectively collecting a name database and a training corpus for language model;

statistically analyzing conventional words and the associated relationships between the conventional words and name words according to the name database and the training corpus for language model; and

generating a basic name language model according to the statistical result.

6. The method according to claim 5, wherein the step of statistically analyzing conventional words and the associated relationships between the conventional words and name words according to the name database and the training corpus for language model comprises:

carrying out name detection in the training corpus according to names in the name database;

replacing all of the specific names in the training corpus with a unified virtual name; and

statistically analyzing conventional words and the associated relationships between the conventional words and name words according to the replaced training corpus.

7. The method according to claim 6, wherein the step of building a basic static decoding network associated with the basic name language model comprises:

providing a virtual pronunciation for the virtual name, so that the virtual name can participates in the static network extension of an acoustic model as a common word;

determining special nodes in the extended static network according to the virtual pronunciation, wherein the special nodes comprise;

a node which enters a name unit and an ending node of a name unit; and

extending a virtual pronunciation unit on an incoming arc or outgoing arc of the special node to obtain a basic static decoding network associated with the basic name language model.

8. The method according to claim 4, wherein the step of generating a name language model of a particular user comprises:

extracting a name from name-associated information associated with the user, and recording the name as a name entry;

setting a word frequency probability for each name entry, and generating a name language model of a particular user according to the word frequency probability of the name entry;

the step of building an affiliated static decoding network associated with the name language model of the particular user comprises;

respectively setting the pronunciation of the word at the beginning and the end of a sentence in the name language model of the particular user to a virtual special pronunciation; and

extending a special pronunciation unit on the outgoing arc of the node at the beginning of a sentence and the incoming arc of the node at the end of a sentence to obtain an affiliated static decoding network associated with the name language model of the particular user.

17. The method according to claim 2, wherein the method further comprises the following steps:

generating a basic name language model and a name language model of a particular user; and

respectively building a basic static decoding network associated with the basic name language model and an affiliated static decoding network associated with the name language model of the particular user.

18. The method according to claim 3, wherein the method further comprises the following steps:

generating a basic name language model and a name language model of a particular user; and

respectively building a basic static decoding network associated with the basic name language model and an affiliated static decoding network associated with the name language model of the particular user.

9. A speech recognition system based on user personalized information, characterized in that comprising:

a receiving unit for receiving a speech signal;

a decoding unit for decoding the speech signal according to a basic static decoding network to obtain a decoding path on each active node in the basic static decoding network, wherein the basic static decoding network is a decoding network formed by extending the words in a basic name language model into corresponding acoustic units, the basic name language model is a language model for describing the statistical probability between common words and the statistical probability between common words and names;

a check unit of decoding path for determining whether a decoding path enters a name node in the basic static decoding network;

a network extending unit for, after the check unit of decoding path determines that a decoding path enters a name node in the basic static decoding network, extending an extra network on the name node according to an affiliated static decoding network of a user, wherein the affiliated static decoding network is a decoding network formed by extending the words in a name language model of a particular user into corresponding acoustic units, the name language model of a particular user is a language model for describing the statistical probability of the specific name associated with the user; and

a result output unit for returning a recognition result after the decoding is completed.

10. The system according to claim 9, wherein the system further comprises:

a determining unit for determining the affiliated static decoding network of the user before the decoding unit decodes the speech signal frame by frame according to the basic static decoding network;

or determining the affiliated static decoding network of the user after the check unit of decoding path determines that a decoding path enters a name node in the basic static decoding network.

11. The system according to claim 10, wherein the determining unit is used specifically for determining the user identity according to a feature of the speech signal, and then determining the affiliated static decoding network of the user according to the user identity;

or determining the user identity according to an equipment code or account number of the user, and then determining the affiliated static decoding network of the user according to the user identity.

12. The system according to claim 9, wherein the system further comprises:

a basic name language model building unit for generating a basic name language model;

a personalized name language model building unit for generating a name language model of a particular user;

a basic static decoding network building unit for building a basic static decoding network associated with the basic name language model; and

an affiliated static decoding network building unit for building an affiliated static decoding network associated with the name language model of the particular user.

13. The system according to claim 12, wherein the basic name language model building unit comprises:

a name collecting unit for collecting a name database;

a corpus collecting unit for collecting training corpus for language model;

a statistics unit for statistically analyzing conventional words and the associated relationships between the conventional words and name words according to the name database and the training corpus for language model; and

a basic name language model building unit for generating a basic name language model according to the statistical result obtained by the statistics unit.

14. The system according to claim 13, wherein the statistics unit comprises:

a detecting subunit for carrying out name detection in the training corpus according to names in the name database;

a replacing subunit for replacing all of the specific names in the training corpus with a unified virtual name; and

a statistics subunit for statistically analyzing conventional words and the associated relationships between the conventional words and name words according to the replaced training corpus.

15. The system according to claim 14, wherein the basic static decoding network building unit comprises:

a virtual pronunciation providing unit for providing a virtual pronunciation for the virtual name, so that the virtual name can participate in the static network extension of an acoustic model as a common word;

a special node determining unit for determining special nodes in the extended static network according to the virtual pronunciation, wherein the special nodes comprise;

a node which enters a name unit and an ending node of a name unit; and

a first extending unit for extending a virtual pronunciation unit on an incoming arc or outgoing arc of the special node to obtain a basic static decoding network associated with the basic name language model.

16. The system according to claim 12, wherein the personalized name language model building unit comprises:

a name extracting unit for extracting a name from name-associated information associated with the user, and recording the name as a name entry;

a personalized name language model building unit for setting a word frequency probability for each name entry, and generating a name language model of a particular user according to the word frequency probability of the name entry;

the affiliated static decoding network building unit comprises;

a setting unit for respectively setting the pronunciation of the word at the beginning and the end of a sentence in the name language model of the particular user to a virtual special pronunciation; and

a second extending unit for extending a special pronunciation unit on the outgoing arc of the node at the beginning of a sentence and the incoming arc of the node at the end of a sentence to obtain an affiliated static decoding network associated with the name language model of the particular user.

19. The system according to claim 10, wherein the system further comprises:

a basic name language model building unit for generating a basic name language model;

a personalized name language model building unit for generating a name language model of a particular user;

a basic static decoding network building unit for building a basic static decoding network associated with the basic name language model; and

an affiliated static decoding network building unit for building an affiliated static decoding network associated with the name language model of the particular user.

20. The system according to claim 11, wherein the system further comprises:

a basic name language model building unit for generating a basic name language model;

a personalized name language model building unit for generating a name language model of a particular user;

a basic static decoding network building unit for building a basic static decoding network associated with the basic name language model; and

an affiliated static decoding network building unit for building an affiliated static decoding network associated with the name language model of the particular user.