一种有效的语音盲信号分离简化混合模型

本文针对语音盲信号分离问题,提出了一种新的简化混合模型.基于从同一声源到距离靠近(距离大约在几十厘米的范围内)的两个麦克风的房间冲激响应具有相似性这一特点,该模型只需利用这两个冲激响应的差异信息,因而简化了盲源分离的计算复杂度.自行设计实现的语音盲信号分离实时DSP实验系统证实,该模型有效可行,特别是通过引入时间延迟来解决非最小相位传递函数的求逆问题使得盲信号分离性能得到显著的改善.

An Effective Simplified Mixing Model in Audio Blind Signal Separation

A novel simplified mixing model (SMM) is proposed for audio blind signal separation (ABSS).The SMM uses the fact that the room impulse responses from a single source to two closely spaced (approximately several 10 cms apart from each other) microphones are very similar.Only the difference between these two impulse responses is needed.As a result,the computational complexity of ABSS is reduced to some extent.A real-time blind signal separation DSP experimental system has been designed and implemented based on the SMM.The real-time experimental results have shown that the proposed simplified mixing model is effective and feasible. In particular,the blind signal separation performance can be improved considerably by introducing a time delay to solve the inverse problem of a non-minimum phase transfer function.