Research on Multi-voice Discourse Structure Analysis and Information Purification Based on Neural Network Technology

Abstract

The overlapping and aliasing of multi-voice signals in complex scenes can easily lead to the blurring of discourse boundaries, the disorder of hierarchical structure and the distortion of key semantic information, which is a key and difficult problem in the field of speech analysis and information processing. Based on the neural network model, this paper constructs a time series feature modeling mechanism for the mixed speech characteristics of multi-person voice, excavates the deep correlation features of speech signals, and accurately disassembles the hierarchical discourse structure of multi-person dialogue. At the same time, through the adaptive redundant interference suppression strategy, the background noise and cross-human voice interference are effectively filtered out, and the accurate purification of the core speech information is completed. The experimental results show that this method can effectively improve the performance of multi-voice analysis and purification in complex aliasing scenarios, and can provide reliable technical reference for related engineering applications such as intelligent speech analysis and speech noise reduction.