PHONOBEST: a noise-robust automatic speech recognition system with phonetic estimation process based on 'expectation'

摘要

We propose a new method, PHONOBEST(PHOnetic kNOwledge Biased ESTimation) for speech recognition in the presence of non-stationary and unpredictable noise by extending PreFEst (Predominant-F0 Estimation Method) that was developed to estimate melody and bass lines from music signals. The proposed method does not need to know about the characteristics of noises in advance and does not even estimate them in its process. A small set of evaluations demonstrate the feasibility of the method by showing good performance even when the background noise is real and non-stationary noise and SNR is 10 dB.