Speech recognition using filter-bank features

Mel-frequency cepstral coefficients (MFCC) have been shown to be very useful in tasks of speech recognition and are the preferred features in state of the art speech recognition systems. We present features derived from filter bank outputs whose performance is comparable to that of MFCCs for connected digit recognition using a Hidden Markov Model (HMM) based speech recognition system. The feature extraction method we present is easily implementable in floating gate analog VLSI circuitry which makes it a viable option for low power speech recognition tasks.