An algorithm for digital time-domain pitch period determination of speech signals and its application to detect F0dynamics in VCV utterances

The algorithm for pitch determination presented in this paper derives pitch periods from speech signals in time domain applying a crude approximation to an inverse filter which is realized as a two-step nonrecursive digital filter. In the first step, a low-pass filter cuts off the higher formants; the second step presents a filter with a zero situated around F1. The output signal of the filter shows a structure similar to the glottal waveform with only one significant peak per pitch period. This peak is detected and marked as a preliminary pitch period limitation. In a final step, these markers are shifted into an optimum phase position. The method permits the realization as a digital on-line piece of hard-ware. The algorithm was applied to detect non-prosodic fundamental frequency dynamics in certain VCV utterances. First results of this experiment show a decrease of f0at the V-C transition which depends on the consonant in the utterance and strongly varies from speaker to speaker.