Multi-person speech separation method and system for chatting robot

Abstract

The invention provides a multi-person speech separation method and system for a chatting robot. The multi-person speech separation method comprises the step that a microphone is used for collecting speech signals to obtain a mixed signal and further comprises the following steps that the mixed signal x is preprocessed; a demixing matrix w is generated randomly; the negative gradient direction of the demixing matrix w is solved through a negative gradient descent method; whether the demixing matrix w is converged or not is judged; the optimal demixing matrix is solved through an improved difference quotient method; an estimation signal of a source signal is solved; and separated speech is output. According to the multi-person speech separation method and system for the chatting robot, the problem of initial value sensitivity of a FastICA algorithm and the problem that the calculation amount is large when multi-person mixed speech is separated are solved, the negative gradient descent method is introduced, thus the initial value sensitivity is overcome, and algorithm convergence stability is improved; and the improved difference quotient method is put forward to replace the FastICA optimizing algorithm, namely a Newton method, and the problem of large calculation amount caused by derivation and Jacobian matrix calculation is avoided.

Description

technical field [0001] The invention relates to the technical field of digital signal processing and computer hearing, in particular to a multi-person voice separation method and system for a chat robot. Background technique [0002] With the rapid development of the Internet and artificial intelligence technology, our way of life and work have changed a lot. The best proof is the wide application of intelligent voice technology in human-computer interaction. In real life, the speech signal of interest is often disturbed by various problems such as ambient background noise from other sources, speech from other speakers, and reverberation from surface reflections. These problems will greatly reduce the intelligibility of speech, and lead to a decline in the performance of subsequent speech recognition and voiceprint recognition. In view of the influence of background noise and other people interfering with speech, it is necessary to perform multi-person mixed speech separati...

AI Technical Summary

Problems solved by technology

The present invention mainly solves the problem of the initial value sensitivity of the FastICA algorithm and the problem of a large amount of calculation when separating multiple people's mixed voices, introduces the negative gradient descent method, overcomes the initial value sensitivity, and enhances the convergence stability of the algorithm; proposes an improved difference quotient method to replace FastICA's optimization algorithm - Newton's method, avoids the problem of large amount of calculation caused by derivation and Jacobian matrix calculation

Images