Auditory-like filterbank: An optimal speech processor for efficient human speech communication

摘要

The transmitter and the receiver in a communication system have to be designed optimally with respect to one another to ensure reliable and efficient communication. Following this principle, we derive an optimal filterbank for processing speech signal in the listener's auditory system (receiver), so that maximum information about the talker's (transmitter) message can be obtained from the filterbank output, leading to efficient communication between the talker and the listener. We consider speech data of 45 talkers from three different languages for designing optimal filterbanks separately for each of them. We find that the computationally derived optimal filterbanks are similar to the empirically established auditory (cochlear) filterbank in the human ear. We also find that the output of the empirically established auditory filterbank provides more than 90% of the maximum information about the talker's message provided by the output of the optimal filterbank. Our experimental findings suggest that the auditory filterbank in human ear functions as a near-optimal speech processor for achieving efficient speech communication between humans.