Analysis-by-synthesis voicing cut-off determination in harmonic coding

In low bit-rate harmonic speech coding, voicing information is often specified by a cut-off frequency of the spectrum. Many approaches of cut-off estimation depend on spectral matching, where a fixed prototype spectrum is used to model voiced harmonics. However, voiced harmonics do not always show a regular shape. One of the causes is harmonic interference. We propose an analysis-by-synthesis voicing cut-off determination scheme that takes into account harmonic interactions in spectral matching. The proposed scheme has been embedded in a 2.4 kb/s harmonic coder. Subjective listening tests show that the scheme performs well and is robust against noise.