A Voice Separation and Tracking Method for Public Security Criminal Investigation Monitoring

Abstract

The invention relates to the technical field of voice signal recognition and processing, and proposes a voice separation and tracking method for public security criminal investigation and monitoring, comprising the following steps: importing initial voice according to time sequence, performing frame-by-frame windowing processing on the initial voice, and obtaining a windowed voice signal Carry out time-frequency decomposition to windowed speech signal, obtain time-frequency two-dimensional signal by short-time Fourier transform; Carry out endpoint detection in frequency domain to described time-frequency two-dimensional signal, the speech signal segment corresponding to empty speech segment Perform filtering processing; use the two-way long-short-term memory network structure to perform speech separation on the time-frequency two-dimensional signal that has completed the filtering process, and output multiple voice waveforms of the target speaker; establish and train a target speaker model based on GMM-UBM, and convert all The speech waveform of the target speaker is used as the model input, and the GMM model of the target speaker is acquired through adaptively, and then the speech waveform is recognized, and the serial number of the target speaker is output, which is the result of speech tracking.

Description

technical field [0001] The invention relates to the technical field of voice signal recognition and processing, and more specifically, to a voice separation and tracking method for public security criminal investigation monitoring. Background technique [0002] In the field of public security criminal investigation and monitoring, it is difficult to obtain relevant important information for the audio clip because the obtained audio clip contains related interference factors such as background noise, multiple speakers, and reverberation. Therefore, in the process of processing the speech signal, it is necessary to separate the speech signals of multiple speakers before processing them separately. At the same time, due to the particularity of criminal investigation monitoring, the voice signals of multiple speakers are collected by the same pickup, so it is difficult to separate and process the voice signals of multiple speakers. In addition, in the actual monitoring process ...

AI Technical Summary

Problems solved by technology

[0004] 1. Align and capture the position information of multiple target speakers through the combination of multiple microphone arrays, but this method has problems of nonlinear combination of multiple microphones and configuration stability;

[0005] 2. Use visual information as auxiliary information to enhance the performance of the speech separation and tracking system to separate and track speech signals. However, this method needs to combine speech information and visual information for simultaneous processing and analysis, and in practical applications There is a delay problem in the collected audio and image, which makes it impossible to adapt;

[0006] 3. The speech signal is processed by using the effective bit coding vector or the speech information of the target speaker as an additional input to the speech separation system, but this method cannot achieve end-to-end speech tracking, and compared with a separate speech tracking algorithm, Due to the introduction of target speaker identity information as input, there is a problem that the time complexity of training and testing is too high

Images