Language recognition method, apparatus and device and computer storage medium
View Patent ↗Embodiments of the present disclosure provide a language recognition method and apparatus, a device and a computer storage medium. In an aspect, in the embodiments of the present disclosure, after the Nth speech segment included by the speech signal is received, language recognition is performed according to already-received previous N speech segments to obtain the score of each language in at least one language, N being 2, 3, 4 . . . ; therefore, if there exists a langue whose score reaches the designated threshold, the language whose score reaches the designated threshold is considered as the language matched with the speech signal. Therefore, the technical solutions according to embodiments of the present disclosure solve the problem in the prior art that the efficiency of language recognition is lower so that the language recognition cannot be applied to an application scenario in which the recognition result needs to be obtained quickly.
1. A language recognition method, wherein the method comprises:
after receiving the Nth speech segment included by a speech signal, performing language recognition according to previous N speech segments that have already been received, to obtain a score of each language in at least one language, N being 2, 3, 4, . . . ; and
if there exists a language whose score reaches a designated threshold, considering the language whose score reaches the designated threshold as a language matched with the speech signal,
wherein the performing language recognition according to previous N speech segments that have already been received, to obtain the score of each language in at least one language comprises:
extracting a speech feature from the Nth speech segment;
obtaining a statistics of the Nth speech segment according to the speech feature;
obtaining a second accumulated statistics according to the statistics of the Nth speech segment and a first accumulated statistics of the previous N−1 speech segments; and
using a language recognition model to perform language recognition for the second accumulated statistics to obtain the score of each language in at least one language.
2. The method according to claim 1 , wherein the method further comprises:
if there does not exist the language whose score reaches the designated threshold, after receiving the N+1th speech segment, continuing to perform language recognition according to the already-received N+1th speech segment, until a language matched with the speech signal is obtained.
3. The method according to claim 1 , wherein before extracting a speech feature from the Nth speech segment, the method further comprises: performing mute detection for the Nth speech segment to remove a mute signal from the Nth speech segment.
4. The method according to claim 1 , wherein the second accumulated statistics comprises: weights of the speech features of previous N speech segments, an average value of speech features of the previous N speech segments, and a variance of the speech features of the previous N speech segments.
5. The method according to claim 1 , wherein the language recognition model comprises a judgment model and a background model; the using the language recognition model to perform language recognition for the second accumulated statistics to obtain the score of each language in at least one language comprises:
using the background model to convert the second accumulated statistics into a corresponding Gaussian hyper vector; and
calculating a space distance between a vector in the judgment model and the Gaussian hyper vector as a score of the language corresponding to the judgment model.
6. The method according to claim 1 , wherein the method further comprises:
if there still does not exist a language whose score reaches the designated threshold after all speech segments included by the speech signal are received, considering a language with the highest score as the language matched with the speech signal.
7. The method according to claim 1 , wherein the method further comprises: considering the language matched with the speech signal as a language recognition result, and outputting the language recognition result.
8. A device, comprising
one or more processor;
a memory;
one or more programs stored in the memory and configured to execute the following operations when executed by the one or more processors:
after receiving the N th speech segment included by a speech signal, performing language recognition according to previous N speech segments that have already been received, to obtain a score of each language in at least one language, N being 2, 3, 4, . . . ; and
if there exists a language whose score reaches a designated threshold, considering the language whose score reaches the designated threshold as a language matched with the speech signal,
wherein the performing language recognition according to previous N speech segments that have already been received, to obtain the score of each language in at least one language comprises:
extracting a speech feature from the Nth speech segment;
obtaining a statistics of the Nth speech segment according to the speech feature;
obtaining a second accumulated statistics according to the statistics of the Nth speech segment and a first accumulated statistics of the previous N−1 speech segments; and
using a language recognition model to perform language recognition for the second accumulated statistics to obtain the score of each language in at least one language.
9. The device according to claim 8 , wherein the operation further comprises:
if there does not exist the language whose score reaches the designated threshold, after receiving the N+1th speech segment, continuing to perform language recognition according to the already-received N+1th speech segment, until a language matched with the speech signal is obtained.
10. The device according to claim 8 , wherein before extracting a speech feature from the Nth speech segment, the operation further comprises: performing mute detection for the Nth speech segment to remove a mute signal from the Nth speech segment.
11. The device according to claim 8 , wherein the second accumulated statistics comprises: weights of the speech features of previous N speech segments, an average value of speech features of the previous N speech segments, and a variance of the speech features of the previous N speech segments.
12. The device according to claim 8 , wherein the language recognition model comprises a judgment model and a background model; the using the language recognition model to perform language recognition for the second accumulated statistics to obtain the score of each language in at least one language comprises:
using the background model to convert the second accumulated statistics into a corresponding Gaussian hyper vector; and
calculating a space distance between a vector in the judgment model and the Gaussian hyper vector as a score of the language corresponding to the judgment model.
13. The device according to claim 8 , wherein the operation further comprises:
if there still does not exist a language whose score reaches the designated threshold after all speech segments included by the speech signal are received, considering a language with the highest score as the language matched with the speech signal.
14. The device according to claim 8 , wherein the operation further comprises: considering the language matched with the speech signal as a language recognition result, and outputting the language recognition result.
15. A computer storage medium which is encoded with a computer program, the program, when executed by one or more computers, enabling said one or more computers to execute the following operations:
after receiving the N th speech segment included by a speech signal, performing language recognition according to previous N speech segments that have already been received, to obtain a score of each language in at least one language, N being 2, 3, 4, . . . ; and
if there exists a language whose score reaches a designated threshold, considering the language whose score reaches the designated threshold as a language matched with the speech signal,
wherein the performing language recognition according to previous N speech segments that have already been received, to obtain the score of each language in at least one language comprises:
extracting a speech feature from the Nth speech segment;
obtaining a statistics of the Nth speech segment according to the speech feature;
obtaining a second accumulated statistics according to the statistics of the Nth speech segment and a first accumulated statistics of the previous N−1 speech segments; and
using a language recognition model to perform language recognition for the second accumulated statistics to obtain the score of each language in at least one language.
16. The computer storage medium according to claim 15 , wherein the operation further comprises:
if there does not exist the language whose score reaches the designated threshold, after receiving the N+1th speech segment, continuing to perform language recognition according to the already-received N+1th speech segment, until a language matched with the speech signal is obtained.
17. The computer storage medium according to claim 15 , wherein before extracting a speech feature from the Nth speech segment, the operation further comprises: performing mute detection for the Nth speech segment to remove a mute signal from the Nth speech segment.
18. The computer storage medium according to claim 15 , wherein the second accumulated statistics comprises: weights of the speech features of previous N speech segments, an average value of speech features of the previous N speech segments, and a variance of the speech features of the previous N speech segments.
19. The computer storage medium according to claim 15 , wherein the language recognition model comprises a judgment model and a background model; the using the language recognition model to perform language recognition for the second accumulated statistics to obtain the score of each language in at least one language comprises:
using the background model to convert the second accumulated statistics into a corresponding Gaussian hyper vector; and
calculating a space distance between a vector in the judgment model and the Gaussian hyper vector as a score of the language corresponding to the judgment model.
20. The computer storage medium according to claim 15 , wherein the operation further comprises:
if there still does not exist a language whose score reaches the designated threshold after all speech segments included by the speech signal are received, considering a language with the highest score as the language matched with the speech signal.
21. The computer storage medium according to claim 15 , wherein the operation further comprises: considering the language matched with the speech signal as a language recognition result, and outputting the language recognition result.