Optimization method, apparatus, device for wake-up model, and storage medium
View Patent ↗Provided are an optimization method, apparatus, device for a wake-up model and a storage medium, which allow for: acquiring a training set and a verification set; performing an iterative training on the wake-up model according to the training set and the verification set; during the iterative training, periodically updating the training set and the verification set according to the wake-up model and a preset corpus database, and continuing performing the iterative training on the wake-up model according to the updated training set and verification set; and outputting the wake-up model when a preset termination condition is reached. The embodiments of the present disclosure, by periodically updating the training set and the verification set according to the wake-up model and the preset corpus database during an iteration, may improve optimization efficiency and effects of the wake-up model, thereby improving stability and adaptability of the wake-up model and avoiding overfitting.
1. An optimization method for a wake-up model, comprising:
acquiring a training set and a verification set;
performing an iterative training on the wake-up model according to the training set and the verification set;
during the iterative training, periodically updating the training set and the verification set according to the wake-up model and a preset corpus database, and continuing performing the iterative training on the wake-up model according to the updated training set and verification set; and
outputting the wake-up model when a preset termination condition is reached;
wherein the periodically updating the training set and the verification set according to the wake-up model and a preset corpus database comprises:
using a preset number of iterations as an iteration period, and inputting a corpus of the preset corpus database into the wake-up model at an end of each iteration period to obtain a wake-up result;
acquiring, according to the wake-up result, a preset number of corpora that generate false wake-ups; and
updating negative example data in the training set and the verification set according to the preset number of corpora that generate false wake-ups.
2. The method according to claim 1 , wherein the acquiring, according to the wake-up result, a preset number of corpora that generate false wake-ups comprises:
acquiring false wake-up probabilities for corpora that generate false wake-ups; and
sorting, according to the false wake-up probabilities, the corpora that generate false wake-ups, and selecting a preset number of corpora that generate false wake-ups with high false wake-up probabilities.
3. The method according to claim 1 , wherein the acquiring a training set and a verification set comprises:
acquiring a Hotword set and a false Hotword set;
respectively acquiring positive example data and negative example data according to the Hotword set and the false Hotword set; and
obtaining the training set and the verification set from the positive example data and the negative example data.
4. The method according to claim 3 , wherein the respectively acquiring positive example data and negative example data according to the Hotword set and the false Hotword set comprises:
performing a data enhancement operation on data in the Hotword set and the false Hotword set, and obtaining the positive example data and the negative example data.
5. The method according to claim 1 , wherein the updating negative example data in the training set and the verification set according to the preset number of corpora that generate false wake-ups comprises:
updating the negative example data in the training set and the verification set after a data enhancement operation is performed on the preset number of corpora that generate false wake-ups.
6. The method according to claim 4 , wherein the data enhancement operation comprises:
superimposing preset noise data and/or reverberation data onto target data to be performed with the data enhancement operation; and/or
adjusting a playback rate of the target data.
7. The method according to claim 5 , wherein the data enhancement operation comprises:
superimposing preset noise data and/or reverberation data onto target data to be performed with the data enhancement operation; and/or
adjusting a playback rate of the target data.
8. An optimization apparatus for a wake-up model, comprising:
a memory, a processor, and a computer program stored on the memory and executable by the processor,
wherein the processor, when running the computer program, is configured to:
acquire a training set and a verification set;
perform an iterative training on the wake-up model according to the training set and the verification set;
during the iterative training, periodically update the training set and the verification set according to the wake-up model and a preset corpus database;
continue performing the iterative training on the wake-up model according to the updated training set and verification set; and
output the wake-up model when a preset termination condition is reached;
wherein the processor is further configured to:
use a preset number of iterations as an iteration period, and input a corpus of the preset corpus database into the wake-up model at an end of each iteration period to obtain a wake-up result;
acquire, according to the wake-up result, a preset number of corpora that generate false wake-ups; and
update negative example data in the training set and the verification set according to the preset number of corpora that generate false wake-ups.
9. The apparatus according to claim 8 , wherein the processor is further configured to:
acquire false wake-up probabilities for corpora that generate false wake-ups; and
sort, according to the false wake-up probabilities, the corpora that generate false wake-ups, and select a preset number of corpora that generate false wake-ups with high false wake-up probabilities.
10. The apparatus according to claim 8 , wherein the processor is further configured to:
acquire a Hotword set and a false Hotword set;
respectively acquire positive example data and negative example data according to the Hotword set and the false Hotword set; and
obtain the training set and the verification set from the positive example data and the negative example data.
11. The apparatus according to claim 10 , wherein the processor is further configured to:
perform a data enhancement operation on data in the Hotword set and the false Hotword set, and obtain the positive example data and the negative example data.
12. The apparatus according to claim 8 , wherein the processor is further configured to:
update the negative example data in the training set and the verification set after a data enhancement operation is performed on the preset number of corpora that generate false wake-ups.
13. The apparatus according to claim 11 , wherein the processor is further configured to:
superimpose preset noise data and/or reverberation data onto target data to be performed with the data enhancement operation; and/or
adjust a playback rate of the target data.
14. The apparatus according to claim 12 , wherein the processor is further configured to:
superimpose preset noise data and/or reverberation data onto target data to be performed with the data enhancement operation; and/or
adjust a playback rate of the target data.
15. A computer readable storage medium having stored thereon a computer program;
wherein the computer program, when being executed by a processor, implements the method according to claim 1 .