Quantitative evaluation of segregated signal with frequency domain binaural model

A study was conducted to compare three methods, such as signal-to-noise ratio (SNR), coherence, and perceptual evaluation of speech quality (PESQ) as the evaluation measures for a segregated signal. These methods were used to discuss the signal segregation performance of frequency domain binaural model (FDBM). SNR and coherence methods readily allowed comparison with other methods, as they were common measures for signal segregation. The estimation of sound source direction by means of FDBM was based on an interaural phase difference (IPD) and an interaural level difference (ILD). Sampling was performed at 16 kHz at 16 bit resolutions and the frame length was set to 512 taps. The hanning window was applied frame by frame and the frame shift was set to 1/4 frame before performing the fast Fourier transform and the inverse transform.