Method and apparatus for recognizing speech

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

determining, in response to receiving a speech input signal, an identity of a target user inputting the speech input signal;

extracting, based on the identity of the target user, a common expression set of the target user from a stored common expression database, the common expression set comprising a plurality of common expressions;

performing acoustic feature extraction on the speech input signal, and inputting an acoustic feature of the speech input signal into an acoustic model to obtain an acoustic model score of the speech input signal;

determining whether a content of the speech input signal is a common expression of the target user, based on the acoustic model score of the speech input signal and acoustic model scores of the common expressions in the stored common expression set of the target user, wherein determining whether the content of the speech input signal is the common expression comprises: determining whether a difference between the acoustic model score of the speech input signal with the acoustic model scores of the common expressions in the stored common expression set of the target user does not satisfy a threshold; and

decoding, in response to the content of the speech input signal is the common expression of the target user, the acoustic feature of the speech input signal using a language model constructed based on common expressions, to obtain a speech recognition result,

wherein the method is performed by at least one hardware processor.

2. The method according to claim 1 , wherein the determining, in response to receiving a speech input signal, an identity of a target user inputting the speech input signal, comprises:

acquiring, in response to receiving the speech input signal, identifier of an electronic device collecting the speech input signal; and

querying the identity of the target user associated with the electronic device based on the identifier of the electronic device collecting the speech input signal.

3. The method according to claim 1 , wherein the determining whether a content of the speech input signal is a common expression of the target user based on the acoustic model score of the speech input signal and acoustic model scores of the common expressions in the stored common expression set of the target user comprises:

determining whether the speech input signal is a noise signal, based on the acoustic model score of the speech input signal; and

comparing, in response to determining the speech input signal is not a noise signal, the acoustic model score of the speech input signal with the acoustic model scores of the common expressions of the target user, and determining whether the content of the speech input signal is the common expression of the target user based on a comparison result.

4. The method according to claim 1 , wherein the decoding, in response to the content of the speech input signal is the common expression of the target user, the acoustic feature of the speech input signal using a language model constructed based on common expressions, to obtain a speech recognition result, comprises:

constructing a state network containing all states in the language model constructed based on common expressions; and

finding a path best matching the acoustic feature of the speech input signal from the state network as the recognition result of the speech input signal.

5. The method according to claim 1 , the method further comprising:

constructing the common expression database, comprising:

collecting historical speech input signals of users, and converting the historical speech input signals into corresponding texts;

clustering historical speech input signals of each of the users based on a consistency between the texts obtained by converting;

counting, based on generation time of the historical speech input signals, generation frequencies of the historical speech input signals in each cluster; and

determining a historical speech input signal in the cluster having a generation frequency greater than a preset frequency as a common expression speech signal of the user, decoding the common expression speech signal of the user to obtain a syllable sequence, forming a corresponding common expression based on the syllable sequence obtained by decoding, and adding the formed common expression to a common expression set of the user.

6. The method according to claim 1 , wherein the acoustic model score of the speech input signal indicates a probability that a frame of the speech input signal corresponds to one or more syllables, and wherein the method further comprises determining that the frame is not a noise signal based on the acoustic model score of the speech input signal.

7. An apparatus for recognizing a speech, the apparatus comprising:

at least one processor; and

a memory storing instructions, the instructions when executed by the at least one processor, cause the at least one processor to perform operations, the operations comprising:

determining, in response to receiving a speech input signal, an identity of a target user inputting the speech input signal;

extracting, based on the identity of the target user, a common expression set of the target user from a stored common expression database, the common expression set comprising a plurality of common expressions;

performing acoustic feature extraction on the speech input signal, and input an acoustic feature of the speech input signal into an acoustic model to obtain an acoustic model score of the speech input signal;

determining whether a content of the speech input signal is a common expression of the target user, based on the acoustic model score of the speech input signal and acoustic model scores of the common expressions in the stored common expression set of the target user, wherein determining whether the content of the speech input signal is the common expression comprises: determining whether a difference between the acoustic model score of the speech input signal with the acoustic model scores of the common expressions in the stored common expression set of the target user does not satisfy a threshold; and

decoding, in response to the content of the speech input signal is the common expression of the target user, the acoustic feature of the speech input signal using a language model constructed based on common expressions, to obtain a speech recognition result.

8. The apparatus according to claim 7 , wherein the determining, in response to receiving a speech input signal, an identity of a target user inputting the speech input signal, comprises:

acquiring, in response to receiving the speech input signal, identifier of an electronic device collecting the speech input signal; and

querying the identity of the target user associated with the electronic device based on the identifier of the electronic device collecting the speech input signal.

9. The apparatus according to claim 7 , wherein the determining whether a content of the speech input signal is a common expression of the target user based on the acoustic model score of the speech input signal and acoustic model scores of the common expressions in the stored common expression set of the target user comprises:

determining whether the speech input signal is a noise signal, based on the acoustic model score of the speech input signal; and

comparing, in response to determining the speech input signal is not a noise signal, the acoustic model score of the speech input signal with the acoustic model scores of the common expressions of the target user, and determine whether the content of the speech input signal is the common expression of the target user based on a comparison result.

10. The apparatus according to claim 7 , wherein the decoding, in response to the content of the speech input signal is the common expression of the target user, the acoustic feature of the speech input signal using a language model constructed based on common expressions, to obtain a speech recognition result, comprises:

constructing a state network containing all states in the language model constructed based on common expressions; and

finding a path best matching the acoustic feature of the speech input signal from the state network as the recognition result of the speech input signal.

11. The apparatus according to claim 7 , wherein the operations further comprise:

constructing the common expression database, comprising:

collecting historical speech input signals of users, and convert the historical speech input signals into corresponding texts;

clustering historical speech input signals of each of the users based on a consistency between the texts obtained by converting;

counting, based on generation time of the historical speech input signals, generation frequencies of the historical speech input signals in each cluster; and

determining a historical speech input signal in the cluster having a generation frequency greater than a preset frequency as a common expression speech signal of the user, decode the common expression speech signal of the user to obtain a syllable sequence, form a corresponding common expression based on the syllable sequence obtained by decoding, and add the formed common expression to a common expression set of the user.

12. A non-transitory computer readable storage medium storing a computer program, wherein the computer program, when executed by a processor, cause the processor to perform operations, the operations comprising:

determining, in response to receiving a speech input signal, an identity of a target user inputting the speech input signal;

extracting, based on the identity of the target user, a common expression set of the target user from a stored common expression database, the common expression set comprising a plurality of common expressions;

performing acoustic feature extraction on the speech input signal, and inputting an acoustic feature of the speech input signal into an acoustic model to obtain an acoustic model score of the speech input signal;

determining whether a content of the speech input signal is a common expression of the target user, based on the acoustic model score of the speech input signal and acoustic model scores of the common expressions in the stored common expression set of the target user, wherein determining whether the content of the speech input signal is the common expression comprises: determining whether a difference between the acoustic model score of the speech input signal with the acoustic model scores of the common expressions in the stored common expression set of the target user does not satisfy a threshold; and

decoding, in response to the content of the speech input signal is the common expression of the target user, the acoustic feature of the speech input signal using a language model constructed based on common expressions, to obtain a speech recognition result.