Speech recognition method and apparatus, and storage medium

1. A speech recognition method, comprising:

acquiring, by a digital signal processor (DSP) of a computing device, audio data;

performing, by the DSP, fuzzy speech recognition on the audio data;

waking up, by the DSP, a central processing unit (CPU) of the computing device in a dormant state upon determining that a wakeup word exists in the audio data based on a fuzzy speech recognition result;

reading, by the CPU, data corresponding to the wakeup word in the audio data from the DSP, to obtain wakeup data;

determining, by the CPU, whether the wakeup word exists in the audio data by performing speech recognition on the wakeup data;

upon determining that the wakeup word exists, performing, by the CPU, semantic analysis on the audio data; and

upon determining that the wakeup word does not exist, determining, by the CPU, that the fuzzy speech recognition result is incorrect and entering the dormant state.

2. The method according to claim 1 , wherein the performing fuzzy speech recognition on the audio data comprises:

performing, by the DSP, speech recognition on the audio data through fuzzy clustering analysis, to obtain the fuzzy speech recognition result.

3. The method according to claim 2 , wherein the performing, by the DSP, speech recognition on the audio data through fuzzy clustering analysis, to obtain the fuzzy speech recognition result comprises:

establishing a fuzzy clustering neural network according to the fuzzy clustering analysis;

predicting, by taking the fuzzy clustering neural network as an estimator of a probability density function, a probability of the audio data comprising the wakeup word;

generating the fuzzy speech recognition result indicating that the wakeup word exists when the predicted probability is greater than or equal to a set value; and

generating the fuzzy speech recognition result indicating that the wakeup word does not exist when the predicted probability is less than the set value.

4. The method according to claim 1 , wherein the performing fuzzy speech recognition on the audio data comprises:

performing, by the DSP, speech recognition on the audio data through a fuzzy matching algorithm, to obtain the fuzzy speech recognition result.

5. The method according to claim 4 , wherein the performing, by the DSP, speech recognition on the audio data through a fuzzy matching algorithm, to obtain the fuzzy speech recognition result comprises:

acquiring a feature map of a pronunciation of the wakeup word, to obtain a standard feature map;

analyzing a feature map of a pronunciation of each word in the audio data, to obtain a to-be-matched feature map;

calculating an extent to which each to-be-matched feature map belongs to the standard feature map according to a membership function;

generating the fuzzy speech recognition result indicating that the wakeup word exists when the extent is greater than or equal to a preset value; and

generating the fuzzy speech recognition result indicating that the wakeup word does not exist when the extent is less than the preset value.

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

after performing the semantic analysis on the audio data, and executing, by the CPU according to an analysis result of the semantic analysis, an operation corresponding to the analysis result.

7. The method according to claim 6 , wherein executing an operation corresponding to the analysis result comprises:

determining an operation object and operation content according to the analysis result; and

executing the operation content on the operation object.

8. The method according to claim 1 , wherein the performing speech recognition on the wakeup data comprises:

setting a working state of the CPU to a first state, the first state being operating with single-core and at low-frequency; and

performing the speech recognition on the wakeup data in the first state.

9. The method according to claim 8 , wherein the performing the semantic analysis on the audio data comprises:

setting the working state of the CPU to a second state, the second state being operating with multi-core and at high-frequency; and

performing the semantic analysis on the audio data in the second state.

10. The method according to claim 1 , wherein the performing the semantic analysis on the audio data comprises:

determining a semantic scene according to the wakeup word corresponding to the audio data;

determining a quantity of working cores and an operation frequency of the CPU according to the semantic scene; and

setting a working state of the CPU according to the quantity of working cores and the operation frequency, to perform the semantic analysis on the audio data.

11. The method according to claim 1 , wherein before the performing fuzzy speech recognition on the audio data, the method further comprises:

performing, by the DSP, noise reduction and/or echo cancellation processing on the audio data.

12. A speech recognition apparatus, comprising:

a digital signal processor (DSP); and

a central processing unit (CPU), wherein:

the DSP is configured to:

acquire audio data;

perform fuzzy speech recognition on the audio data; and

wake up the CPU in a dormant state upon determining that a wakeup word exists in the audio data based on a fuzzy speech recognition result; and

the CPU is configured to:

read data corresponding to the wakeup word in the audio data from the DSP, to obtain wakeup data;

determine whether the wakeup word exists in the audio data by performing speech recognition on the wakeup data;

upon determining that the wakeup word exists, perform semantic analysis on the audio data; and

upon determining that the wakeup word does not exist, determine that the fuzzy speech recognition result is incorrect and enter the dormant state.

13. The apparatus according to claim 12 , wherein the CPU is further configured to:

after performing the semantic analysis on the audio data, execute an operation according to an analysis result of the semantic analysis.

14. The apparatus according to claim 12 , wherein the CPU is further configured to:

perform the speech recognition on the wakeup data at a first working state of single-core and low-frequency operation.

15. The apparatus according to claim 14 , wherein the CPU is further configured to:

perform the semantic analysis on the audio data at a second working state of multi-core and high-frequency operation.

16. The apparatus according to claim 12 , wherein the CPU is further configured to:

determine a semantic scene according to the wakeup word corresponding to the audio data;

determine a quantity of working cores and an operation frequency of the CPU according to the semantic scene; and

perform the semantic analysis on the audio data in a third working state according to the quantity of working cores and the operation frequency.

17. A non-transitory computer readable storage medium storing a plurality of instructions, the instructions, when being executed by one or more processors of a computing device, cause the one or more processors to perform:

acquiring, by a digital signal processor (DSP) of the computing device, audio data;

performing, by the DSP, fuzzy speech recognition on the audio data;

waking up, by the DSP, a central processing unit (CPU) of the computing device in a dormant state upon determining that a wakeup word exists in the audio data based on a fuzzy speech recognition result;

reading, by the CPU, data corresponding to the wakeup word in the audio data from the DSP, to obtain wakeup data;

determining, by the CPU, whether the wakeup word exists in the audio data by performing speech recognition on the wakeup data;

upon determining that the wakeup word exists, performing, by the CPU, semantic analysis on the audio data; and

upon determining that the wakeup word does not exist, determining, by the CPU, that the fuzzy speech recognition result is incorrect and entering the dormant state.

18. The storage medium according to claim 17 , wherein the instructions further cause the CPU to:

perform the speech recognition on the wakeup data at a first working state of single-core and low-frequency operation.

19. The storage medium according to claim 18 , wherein the instructions further cause the CPU to:

perform the semantic analysis on the audio data at a second working state of multi-core and high-frequency operation.

20. The storage medium according to claim 17 , wherein the instructions further cause the CPU to:

determine a semantic scene according to the wakeup word corresponding to the audio data;

determine a quantity of working cores and an operation frequency of the CPU according to the semantic scene; and

perform the semantic analysis on the audio data in a third working state according to the quantity of working cores and the operation frequency.