Source waveform recovery under reverberant conditions is quite important for machine diagnostics and speech quality enhancement. His paper investigates power envelope dereverberation without using reverberant transfer functions. Source waveforms and reverberant waveforms are typically modeled here by modulation transfer functions (MTF). Using this signal modeling, we theoretically showed that the reverberant power envelope was given by the convolution of the source and impulse response power envelopes. The dereverberation experiments demonstrate that MTF and speeds signals are recoverd from power envelope inverse filtering.
展开▼
