Use of computational psychoacoustical models in speech processing: Coding and objective performance evaluation

The primary objective of this study was to determine what benefit could be gained in speech coding by using a psychoacoustical frequency scale instead of a linear scale. To partially overcome the well known difficulties in objective speech quality measurements, a computational performance criterium based on psychoacoustical models was developed. Several Finnish phonemes were then coded using regular LPC and LPC computed on a psychoacoustically correct frequency scale (Bark scale)and the coding performance of these both methods was tested via computational performance tests. The results indicate a significant improvement in speech quality for the same bit-rate, when applying LPC on psycho-acoustical frequency scale. Preliminary listening tests support both the better coding capability of the Bark LPC compared to the regular LPC and the reliability of the developed speech quality criterium as an objective performance evaluation method.