Both for robust fundamental frequency (F0) measurement and to provide a degree of voicing indication, a new algorithm has been developed based on multi-channel autocorrelation analysis. The speech is filtered into eight separate frequency bands, representing the lowest 500 Hz and seven overlapping band-pass channels each about 1000 Hz wide. The outputs of all the band-pass channels are full-wave rectified and band-pass filtered between 50 Hz and 500 Hz. Autocorrelation functions are calculated for the signals from all eight channels, and these functions are used both for the F0 measurement and for the voicing indication. Optional dynamic programming is provided to maximize the continuity of position of the correlation peaks selected for fundamental period measurement. The algorithm has been designed for coding onto a 16-bit integer DSP, using less than 4 MIPS processing power and 1500 words of data memory.
Cite as: Holmes, J.N. (1998) Robust measurement of fundamental frequency and degree of voicing. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0351, doi: 10.21437/ICSLP.1998-525
@inproceedings{holmes98b_icslp,
author={John N. Holmes},
title={{Robust measurement of fundamental frequency and degree of voicing}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 0351},
doi={10.21437/ICSLP.1998-525}
}