Speech recognition method and system based on user personalized information

1. A method, comprising:

receiving, by a processing device, a speech signal;

decoding, by the processing device, 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 formed by extending words in a basic name language model into corresponding acoustic units, and wherein the basic name language model comprises a first statistical probability between two common words and a second statistical probability between a common word and a name;

generating, by the processing device, a user-specific name language model comprising a third statistical probability between the name and a user identifier;

building, by the processing device, an affiliated static decoding network associated with the user-specific name language model by extending words in the user-specific name language model into corresponding acoustic units, wherein building the affiliated static decoding network further comprises:

setting a first pronunciation of a first word at a beginning of a sentence in the user-specific name language model to a first virtual pronunciation;

setting a second pronunciation of a second word at an end of the sentence in the user-specific name language model to a second virtual pronunciation; and

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

responsive to identifying that a decoding path enters a name node in the basic static decoding network, extending, by the processing device, an extra network associated with the name node according to the affiliated static decoding network; and

returning, by the processing device, a recognition result after the decoding is completed, wherein a recognition accuracy rate for names is improved.

2. The method according to claim 1 , further comprising:

determining the affiliated static decoding network one of before or after the speech signal is decoded according to the basic static decoding network.

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

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

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

4. The method according to claim 1 , further comprising:

generating the basic name language model; and

building the basic static decoding network associated with the basic name language model.

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

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

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

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

6. The method according to claim 5 , wherein 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 further comprises:

performing name detection in the training corpus according to names in the name database;

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

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

7. The method according to claim 6 , wherein generating the basic static decoding network further comprises:

providing a virtual pronunciation for the virtual name to allow the virtual name participating in 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 the name unit; and

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

8. The method according to claim 4 , wherein generating the user-specific name language model further comprises:

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

setting a word frequency probability for the name entry; and

generating the user-specific name language model according to a word frequency probability of the name entry.

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

a memory; and

a processing device, communicatively coupled to the memory, to:

receive a speech signal;

decode 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 formed by extending words in a basic name language model into corresponding acoustic units, and wherein the basic name language model comprises a first statistical probability between two common words and a second statistical probability between a common word and a name;

generate a user-specific name language model comprising a third statistical probability between the name and a user identifier;

build an affiliated static decoding network associated with the user-specific name language model by extending words in the user-specific name language model into corresponding acoustic units, wherein to build the affiliated static decoding network, the processing device is further to:

set a first pronunciation of a first word at a beginning of a sentence in the user-specific name language model to a first virtual pronunciation;

set a second pronunciation of a second word at an end of the sentence in the user-specific name language model to a second virtual pronunciation; and

extend a special pronunciation unit on an outgoing arc of a node corresponding to the beginning of the sentence and an incoming arc of the node corresponding to the end of the sentence to obtain the affiliated static decoding network associated with the user-specific name language model; and

responsive to identifying that a decoding path enters a name node in the basic static decoding network, extend an extra network associated with the name node according to the affiliated static decoding network; and

return a recognition result after the decoding is completed, wherein a recognition accuracy rate for names is improved.

10. The system according to claim 9 , wherein to determine the affiliated static decoding network, the processing device is further to determining the affiliated static decoding network one of before or after the speech signal is decoded according to the basic static decoding network.

11. The system according to claim 10 , wherein to determine the affiliated static decoding network, the processing device is further to:

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

determine the identity of the user according to at least one of an equipment code or an account number associated with the user, and then determining the affiliated static decoding network according to the identity of the user.

12. The system according to claim 9 , wherein the processing device is further to:

generate the basic name language model; and

build the basic static decoding network associated with the basic name language model.

13. The system according to claim 12 , wherein to generate the basic name language model, the processing device is further to:

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

statistically analyze conventional words and associated relationships between the conventional words and name words according to the name database and the training corpus for the language model to generate a statistical result; and

generate the basic name language model according to the statistical result.

14. The system according to claim 13 , wherein to analyze conventional words and the associated relationships, the processing device is further to:

perform name detection in the training corpus according to names in the name database;

replace all specific names in the training corpus with a unified virtual name to update the training corpus; and

statistically analyze conventional words and the associated relationships between the conventional words and name words according to the updated training corpus.

15. The system according to claim 14 , wherein to generate the basic static decoding network, the processing device is further to:

provide a virtual pronunciation for the virtual name to allow the virtual name participating in static network extension of an acoustic model as a common word;

determine 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 the name unit; and

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

16. The system according to claim 12 , wherein to generate the user-specific name language model, the processing device is further to:

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

set a word frequency probability for the name entry; and

generate the user-specific name language model according to a word frequency probability of the name entry.