Frequency domain binaural model using interaural level and phase differences and its application for signal segregation

摘要

We can communicate others under noisy environments. This phenomenon is due to the binaural function and called as a cock-tail party effect. This paper proposes the frequency domain binaural model (FDBM) using interaural level and phase differences. The performance of the proposed FDBM is confirmed by speech segregation and speech recognition task. As a result of the speech recognition task, more than 90% recognition rate is obtained within a pink noise for SNR of 5dB condition. And the ability for FDBM as a speech enhancer is also shown by waveform and spectrogram of the segregated signal.