Speech recognition method and device

Abstract

The invention discloses a speech recognition method and device, and relates to the technical field of computers. A specific embodiment of the method comprises the steps of extracting pre-training features corresponding to an unlabeled first audio data sample through a feature extraction network, and obtaining a normalized weight vector of phonemes of the first audio data sample through a feature mapping network based on the pre-training features; and taking the normalized weight vector as a training target corresponding to the first audio data sample and taking a label of a labeled second audio data sample as a training target corresponding to the second audio data sample, training a speech recognition model, and performing speech recognition by using the trained speech recognition model. According to the implementation mode, the problems of data dependence and voice representation of voice recognition can be solved, the voice recognition performance is improved by effectively utilizing unmarked audio data in a voice recognition product, the manual marking cost is reduced, and the problems that in the prior art, voice phase information is ignored, and complex voice characteristic modeling capacity has defects are solved.

Description

technical field [0001] The present invention relates to the field of computer technology, and in particular, to a speech recognition method and device. Background technique [0002] Speech recognition technology is designed to solve the conversion problem from speech audio signal to speech text. Based on the results of speech recognition, the integration of natural language understanding, multi-modal fusion and other technologies can achieve the purpose of human-computer interaction. The current speech recognition system usually adopts a supervised training scheme, that is, based on manual annotation of the collected audio data, according to the original audio data and features, with text annotation as the final goal, a classifier for speech recognition is trained. There are two types of speech recognition technologies in common use today. A kind of hybrid framework based on Hidden Markov Deep Neural Network (HMM-DNN), which is divided into two modules: acoustic model and ...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the embodiments of the present invention provide a speech recognition method and device, which can solve the data dependence and speech representation problems of speech recognition in various business fields and application scenarios, and can effectively utilize a large number of unlabeled speech recognition products in existing speech recognition products. Audio data is used to improve the performance of speech recognition, reduce the cost of manual labeling, reduce the time-consuming labeling, and improve the accuracy of labeling. It is suitable for super-large-scale speech recognition training, and solves the problem of ignoring voice phase information and modeling complex voice characteristics in existing technologies. flawed problem

Images