Abstract:
A robust echo canceller using an STFT-CS (short-time Fourier transform and cross spectrum) method under noisy conditions was proposed [J. Acoust. Soc. Am. 99, 2505(A) (1996)]. This method is developed by using an algorithm which estimates the room transfer function (RTF) using a frame average of the cross spectrum between the input and error signals. Taking the cross-spectrum frame average reduces the RTF estimation errors produced by surrounding noise. The robustness and convergence properties of STFT-CS and other conventional methods under noisy conditions are investigated. STFT-CS is robust to surrounding noise, but it does not rapidly follow temporarily changing RTFs due to taking the frame average. Scattering waves from listeners (and/or talkers) cause significant changes in the RTF. Thus the scattering sound fields from a sphere are investigated as an initial step to develop an RTF prediction method for STFT-CS under noisy conditions. The time waveforms and frequency characteristics of the scattering waves are analyzed numerically, and the RTF prediction errors are evaluated experimentally for STFT-CS and conventional methods under noisy conditions.