A linear predictive vocoder with new pitch extraction and exciting source

We developed an experimental model of linear predictive vocoder that transmits speech signals at 2,400 bps. This equipment employs a nonlinear pre-processing for pitch extraction which is particulary efficient for band-limited speech signals. The pitch period is searched adaptively upon the values so that no double/half pitch period could be detected. The voiced/unvoiced decision is made based on the maximum normalized value of autocorrelation function, the total energy and the ratio of residual energy to the total one. In the synthesis part the composite pulse is supplied as an exciting source during the voiced sound period. The subjective evaluation tests indicate that the syllable articulation of about 66 % can be attainable in Japanese.