<P>Reverberation causes a performance degradation in distinct speech processing. For this reason, quantitatively estimating the amount of reverberation from the signal received by the microphone has been an important task for characterizing room...
http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
https://www.riss.kr/link?id=A107456151
2018
-
학술저널
486-495(10쪽)
0
상세조회0
다운로드다국어 초록 (Multilingual Abstract)
<P>Reverberation causes a performance degradation in distinct speech processing. For this reason, quantitatively estimating the amount of reverberation from the signal received by the microphone has been an important task for characterizing room...
<P>Reverberation causes a performance degradation in distinct speech processing. For this reason, quantitatively estimating the amount of reverberation from the signal received by the microphone has been an important task for characterizing room acoustics and compensating for degradation due to an algorithm. In this paper, a novel method that estimates the reverberation time (T-60) based on multi-channel microphones using a deep neural network (DNN) is proposed. Each channel's distribution of the decay rates for each frequency and the generalized cross-correlation with phase transform (GCC-PHAT) between the microphones are adopted as the input feature vectors for DNN training. Those refined features enable the DNN composed of multiple nonlinear hidden layers to learn the nonlinear relationship that labels the reverberation time from the input features, which is known to be challenging with low-order features. The proposed algorithm is evaluated with extensive noisy conditions, and the results show the advantage of employing multi-channel signals with spatial features when compared with conventional methods.</P>