The noise filter, same as implemented in sphinxbase/sphinxtrain/pocketsphinx.
Noise removal algorithm is inspired by the following papers Computationally
Efficient Speech Enchancement by Spectral Minina Tracking by G. Doblinger
Power-Normalized Cepstral Coefficients (PNCC) for Robust Speech Recognition
by C. Kim.
For the recent research and state of art see papers about IMRCA and A
Minimum-Mean-Square-Error Noise Reduction Algorithm On Mel-Frequency Cepstra
For Robust Speech Recognition by Dong Yu and others