EUROSPEECH '89
First European Conference on Speech Communication and Technology

Paris, France
September 27-29, 1989

A Fast Expert Program for Pitch Extraction

Henning Reetz

Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands

A simple and fast algorithm is presented, which extracts international contours over long speech segments with laryngograph-like quality. The algorithm operates in the time domain, is resistant against (non-periodic) noise, and detects pitch reliable in a range between 50 Hz and 1000 Hz without parameter adjustments or voice/voiceless pre-segmenting. The algorithm is divided into four major steps. First, a speech signal is reduced to positive and negative peaks that represent (possible) glottal pulses. Second, the amplitude and frequency variation of the peak train is reduced. Third, a way with minimal frequency variation through the remaining peaks is searched. The distances between the peaks of this way give the duration of the pitch periods. Finally, segments that are too short, and segments with a pitch higher or lower than a given range are eliminated. The algorithm has proven to give reliable results in all applications. The computing time of the algorithm is less than realtime on a VAX 750.

Full Paper

Bibliographic reference.  Reetz, Henning (1989): "A fast expert program for pitch extraction", In EUROSPEECH-1989, 1476-1479.