Variational Bayes Estimation Approach to Multi-microphone Speech Dereverberation

STFT coefficients of are modeled as Gaussian random variables with time-dependent variance. The variance, being a positive quantity is modeled as a Gamma distributed random variable.


Speech Examples

Two microphone scenario

MCLP Order: 16
Male speech Female speech
Clean speech:
Reference Microphone signal:
Enhanced signal:
WPE:
CGG:
VBE:
Clean speech:
Reference Microphone signal:
Enhanced signal:
WPE:
CGG:
VBE:

Four microphone scenario

MCLP Order: 16
Male speech Female speech
Clean speech:
Reference Microphone signal:
Enhanced signal:
WPE:
CGG:
VBE:
Clean speech:
Reference Microphone signal:
Enhanced signal:
WPE:
CGG:
VBE:

Eight microphone scenario

MCLP Order: 8
Male speech Female speech
Clean speech:
Reference Microphone signal:
Enhanced signal:
WPE:
CGG:
VBE:
Clean speech:
Reference Microphone signal:
Enhanced signal:
WPE:
CGG:
VBE: