Speech recognition post-processing method, device and computer equipment

Abstract

The present application provides a speech recognition post-processing method, device and computer equipment. The recognition system first obtains a sequence matrix output by a speech recognition model and a preset candidate word list, wherein the candidate word list contains multiple candidate words. The recognition system calculates the candidate word scores corresponding to each candidate word according to the sum of candidate word paths of each candidate word in the sequence matrix. Finally, the control system screens and obtains the final result of the speech recognition according to the candidate word scores and candidate word lengths corresponding to each candidate word. The present application filters the sequence matrix on the basis of the preset candidate word list, thereby reducing the search scale of the candidate word path and improving the search speed. And when screening the final results of speech recognition, the candidate word scores and candidate word lengths of each candidate word are integrated, thereby avoiding the shortcoming that the recognition results are biased toward shorter candidate words, and effectively improving the accuracy of speech recognition.

Description

technical field [0001] The present application relates to the technical field of speech recognition, in particular to a speech recognition post-processing method, device and computer equipment. Background technique [0002] Recognition of wake-up words and command words belongs to the field of speech recognition and is widely used in scenarios such as smart homes and smart terminals. When the existing wake-up word and command word models are applied, the user's voice is detected in real time, and when specific words are detected, feedback is given. The general post-processing method of speech recognition is: search for the optimal path in the decoding space output by the speech recognition model (for example, the CTC method is the phoneme probability matrix output by the neural network, and the size is the matrix of the sequence length * the number of phonemes). is the final recognition result. [0003] When searching for the optimal path in the decoding space, the basic p...

AI Technical Summary

Problems solved by technology

[0004] The main purpose of this application is to provide a post-processing method, device and computer equipment for speech recognition, aiming to solve the disadvantages of low accuracy of existing speech recognition

Images