Multi-user dialogue audio recognition method and system based on machine learning

Abstract

The invention provides a multi-user dialogue audio role recognition dialogue audio method based on machine learning. The method specifically includes steps: training voice data with marks by employing a UBM-GMM algorithm to obtain a UBM-GMM model; performing secondary segmentation on to-be-recognized voice data, and clustering voice to obtain voice samples with clustering marks; then extracting a part of the voice samples with the clustering marks as samples to be put into the UBM-GMM model, and performing alignment training to obtain an alignment training model; performing identity recognition according to the alignment training model to obtain collected voice clips with the same identity, and classifying the voices with the same identity; and outputting the voice data of each person in a dialogue. According to the method, roles in the audio are intelligently sampled for the role recognition model training, the precision of voice segmentation and role restoration is greatly improved, automatic optimization of the model is realized, and problems of low segmentation and recognition precision and failure of dynamic optimization of the conventional method are solved.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a machine learning-based multi-person dialogue audio recognition method and system. Background technique [0002] Dialogue audio contains two or more character dialogues, from which it is very important to identify and extract the words spoken by each character and convert them into text dialogues, which is of great significance for the in-depth analysis and application of audio content. [0003] Existing dialogue segmentation techniques are mainly based on statistical distance methods, such as Bayesian Information Criterion (BIC) and Generalize Likelihood Ratio (GLR). The method based on statistical distance has the advantage that it does not need to go through the step of training the model with sample data, and directly assumes that there are differences in the distribution of the Gaussian Mixture Model (GMM) of the acoustic model of different people in a short perio...

AI Technical Summary

Problems solved by technology

The method based on statistical distance has the advantage that it does not need to go through the step of training the model with sample data, and directly assumes that there are differences in the distribution of the Gaussian Mixture Model (GMM) of the acoustic model of different people in a short period of time, and then based on this difference, it can be segmented. This method can be applied to any voice role segmentation task; however, the method based on statistical distance has obvious disadvantages. It is a general segmentation method that does not consider personalized scenes, and does not consider the problem of character recognition. The accuracy of segmentation and character recognition is relatively low. Difference

Images