Voice recognition method and system

Abstract

The invention relates to the field of voice recognition, and discloses a voice recognition method and system. The method comprises the following steps: carrying out the spectrum analysis of the obtained N voice signals, and obtaining multi-dimensional N preliminary frequency spectrum characteristic parameters, inputting the parameters into N samples of a pre-trained auditory perception model based on a deep neural network for characteristic transformation, and obtaining N refined auditory perception characteristics; enabling the combination of the N refined auditory perception characteristics to be inputted into a pre-trained acoustic classification model, and coding the output so as to recognize a text content corresponding to a voice signal. According to the invention, the method carries out the frequency spectrum analysis and the characteristic transformation of the obtained multipath voice signals, and achieves the supplementary effect for the auditory perception. The extracted auditory perception characteristics are more suitable for the auditory perception of human ears, and the method can improve the voice recognition accuracy and anti-noise capability.

Description

technical field [0001] The invention relates to the field of pattern recognition, in particular to the technical field of speech recognition. Background technique [0002] Speech is the acoustic performance of language, the most natural, effective and convenient means for human to exchange information, and also a kind of support for human thinking. In the era of mobile Internet, speech recognition is one of the very important human-computer interaction technologies. In today's information society and fast-paced life, using signal processing and pattern recognition technology makes it possible to use machines for automatic speech recognition, which is important for Improving production efficiency and quality of life is of great significance. Automatic speech recognition has a wide range of applications. It can turn handwritten documents into automatic dictation operations, use voice to remotely control home appliances, use voice to search for events of interest on the Intern...

AI Technical Summary

Problems solved by technology

[0005] In recent years, with the development of signal processing and machine learning, speech recognition research has achieved great success, including Gaussian mixture model (Gaussion mixture model, referred to as GMM), hidden Markov model (Hidden markov model, referred to as Methods including HMM) and deep neural networks have achieved high recognition accuracy, but the performance in noisy environments is still not satisfactory, and the effects of existing algorithms for different noises vary greatly.

[0006] Therefore, it is necessary to solve the problem of performance degradation of the existing speech recognition system in a noisy environment, in order to improve the applicability and practicability of the speech recognition system, and try to approach and achieve the ability of human ear speech perception

Images