Whispered speech recognition based on gammatone filterbank cepstral coefficients