Method and device for training mixed language recognition model

Abstract

The embodiment of the present application discloses a method and device for training a mixed language recognition model. A specific embodiment of the method includes: generating the first syllable label sequence of the first language audio and the second syllable label sequence of the second language audio; using the first language recognition model to obtain the first connection time series classification Viterbi sequence and the second Linkage timing sorting Viterbi sequences; determining linking timing sorting Viterbi scores for each first syllable label and linking timing for each second syllable label based on the first linking timing sorting Viterbi sequence and the second linking timing sorting Viterbi sequence Categorical Viterbi score; Differential syllable labels from the second syllable label sequence based on the difference in connected temporal categorical Viterbi scores; Hybrid training of deep neural networks based on the first syllable label sequence and differential syllable labels to obtain a mixed language recognition model . This embodiment realizes that the same model supports recognition of multiple languages.

Description

technical field [0001] The embodiments of the present application relate to the field of computer technology, and in particular to a method and device for training a mixed language recognition model. Background technique [0002] With the development of speech recognition technology, the performance of speech recognition has been practical. For example, various input methods on mobile phones have voice interaction functions. In practical applications, in addition to the speech recognition of the Mandarin scene, there is also the speech recognition of the dialect scene. At present, there are many voice interaction products that support dialect voice recognition, such as voice recognition options on mobile phone input methods, users can choose the corresponding dialect according to their needs, and some smart TVs and smart refrigerators designed for specific dialects. [0003] In related technologies, a Mandarin recognition model is usually used for speech recognition of Mand...

AI Technical Summary

Problems solved by technology

[0003] In related technologies, the Mandarin recognition model is usually used for speech recognition of Mandarin, and the corresponding dialect recognition model is used for speech recognition of dialects. However, when users switch languages, they need to select the corresponding speech recognition model back and forth, which is cumbersome to operate.

Moreover, as more and more dialects are supported, more and more dialect recognition models need to be trained, resulting in a higher workload for model training

Images