Investigating Speech Quality by Homomorphic Deconvolution

The quality of speech depends on how the speech is processed before it reaches the ears of the perceiving person. Considered in this work is the most common kind of processing the transmission of sound in a closed room. Room acoustics play an important role in the assessment of speech quality. The distance between two points in the room can be regarded as a digital transfer function. We subject such functions to cepstral domain analysis revealing that the speech quality decreases with the amount of excess phase in the transfer function. This result emerges from a controlled listening test where ten test-persons were exposed to a speech signal filtered by different transfer functions. Since an excess phase function is not invertible, it is not possible to compensate for the quality degradation due to that part of a room transfer function.