Noise-robust hands-free speech recognition using SIMO-model-based blind source separation

This paper addresses a noise-robust hands-free speech recognition system in which Single-Input Multiple-Output (SIMO)-model-based blind source separation (BSS) is introduced to achieve a fluent human-machine interaction. The method consists of SIMO-ICA technique and SIMO-model-based binary masking, which can achieve a superior noise reduction with low distortion. In this paper, we investigate the distortion controllability of the two-stage BSS method. The experimental results reveal that the speech recognition rate can be considerably improved by using the two-stage method compared with the conventional BSS methods. In addition, a BSS-based hands-free spoken dialogue system is illustrated.