Speech/noise-dominant decision regardless of SNR for speech enhancemen
Yukihiro Nomura, Jianming Lu, Hiroo Sekiya, and Takashi Yahagi
2004 RISP International Workshop on Nonlinear Circuits and Signal Processing(NCSP2004), pp.415-418, Mar., 2004. [pdf document]

<Abstract>

In speech enhancement, a decision between speech dominant and noise one is important to reduce noise for increasing intelligibility. This paper presents a speech/noise-dominant decision regardless of SNR. In the proposed decision, the influence of noise is reduced by subtracting the noise component. Therefore, the proposed method decides between the speech dominant and noise one accurately. From the investigation of segmental SNR, Itakura-Saito istortion measure and listening tests, the speech enhancement using the proposed decision reduces speech distortion though segmental SNR improvement is kept.