Speech recognition method and system

Abstract

The invention discloses a voice recognition method and system, and relates to the technical field of voice recognition. The method comprises the following steps: constructing a speech recognition model based on a Transform model and a WFST model; acquiring a to-be-recognized audio signal; detecting a voice signal in the to-be-recognized audio signal to obtain a target voice signal; performing transformation processing on the target voice signal to obtain a voice feature vector sequence; performing transformation processing on the voice feature vector sequence to obtain a target voice feature sequence; inputting the target speech feature sequence into a speech recognition model; recognizing the target speech feature sequence through a Transform model in the speech recognition model, and outputting a phoneme sequence; and inputting the phoneme sequence into a WFST model in the speech recognition model, and outputting a Chinese character sequence to complete speech recognition. According to the invention, the data purchase cost can be effectively reduced, and the accuracy of speech recognition is ensured.

Description

technical field [0001] The present invention relates to the technical field of voice recognition, in particular, to a voice recognition method and system. Background technique [0002] Speech recognition is essentially the process of converting an audio sequence to a text sequence, that is, to find the text sequence with the highest probability given a speech input. Based on the Bayesian principle, the speech recognition problem can be decomposed into the conditional probability of the occurrence of this speech in a given text sequence and the prior probability of the occurrence of the text sequence. The model obtained by modeling the conditional probability is the acoustic model. The model obtained by modeling the prior probability of the text sequence is the language model. [0003] An acoustic model is the output of converting speech into an acoustic representation, that is, finding the probability that a given speech originates from an acoustic symbol. For acoustic sym...

AI Technical Summary

Problems solved by technology

Among the above-mentioned technologies, the former one treats the model as a black box, trains the speech and text annotations, and directly obtains the text by inputting the speech. The disadvantage of this technology is that the model training requires very rich speech data and annotations. Corresponding to extremely high data procurement costs

The latter technique is to strictly fill in the training data and parameters according to the regulations of the framework, and the intermediate links are very tightly coupled, so the disadvantage is that the framework is difficult to modify, which limits the integration of other advanced models

Images