In this work, we assume reverberant environment. Reverberation
is created in enclosed spaces when acoustic waves
reflected by the walls and objects arrive to the microphone
attenuated and with different delays, introducing undesirable
and unpredictable interferences. The reverberation distortion
is usually modeled through a convolution between
the room impulse response (RIR) and the clean speech signal.
That linear distortion from the acoustic channel can
not be easily canceled or attenuated in the feature domain
as it is routinely done in ASR for the linear distortions produced
in the electric channel (microphone, amplifiers, telephone
network, etc.), since the duration of the RIR is
usually much longer than the electrical channel impulse
responses and encompasses several consecutive phones