Voice recognition method and device, computer equipment and storage medium

Abstract

The invention discloses a voice recognition method and device, computer equipment and a storage medium. The method comprises the following steps of: acquiring spectrum feature data of a voice signal to be recognized; and inputting the spectrum feature data into a pre-trained voice recognition model, and recognizing the category of the voice signal by the voice recognition model, the voice recognition model being used for extracting sequence features from the spectrum feature data and performing category recognition based on the extracted sequence features, and the voice recognition model being obtained by training a voice sample marked with a voice category. According to the method, the spectrum feature data are used as input features, a plurality of deep learning neural network structures are fused, the feature extraction capability of the model on a time sequence is enhanced, and the voice classification efficiency is improved.

Description

technical field [0001] The present invention relates to the technical field of deep learning, in particular to a voice recognition method, device, computer equipment and storage medium. Background technique [0002] In some infectious diseases, coughing can not only effectively remind the observed subjects whether they may be infected, but also help to make an effective judgment on the development of the disease. At present, the sound recording technology is relatively mature, but the content of the recording still needs to be analyzed by manpower, and the workload of selecting the clip where the cough occurs is very heavy. If computer-aided identification of cough signals can be used, it will be of great significance for doctors to diagnose and treat patients. [0003] At present, cough sound recognition methods mainly include dynamic time warping (DTW), clustering, hidden Markov process (HMM) and the combination and improvement of various algorithms. After manually extra...

AI Technical Summary

Problems solved by technology

[0005] However, there may be problems such as loud noise, inconspicuous cough, variable recording distance, and different cough characteristics of different people in the recording, which brings great technical challenges to the cough recognition model.

For example, the spectrum obtained by a short-distance cough and a long-distance cough will be significantly different, and it is easy to make mistakes when performing feature calculations; a piece of audio coughing at different time points will also cause a large difference in the feature vector; what's more, When there are sharp noises or metal impact sounds in the sound, the sound difference is not high, and the classification accuracy of the traditional feature vector represented by Mel cepstral coefficient (MFCC) is not high.

Images