Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

A Cepstrum-Based Harmonics-To-Noise Ratio in Voice Signals

Peter J. Murphy

Department of Electronic and Computer Engineering, University of Limerick, National Technological Park, Plassey, Limerick, Ireland

A new cepstrum-based technique is developed in order to provide an alternative means of estimating the harmonics-to-noise ratio in voice signals. The geometric mean harmonics-to-noise ratio (GHNR) is defined as the mean of the individual spectral (i.e. at specific frequency locations) harmonics-to-noise ratios in dB. A heuristic development of the method treats the harmonic spectrum (in dB) of voiced speech taken over several cycles of the waveform as a more usual time domain signal, which is Fourier transformed. The sum of the resulting cepstral peaks (rahmonics) gives a direct estimation of the geometric mean harmonics-to-noise ratio (GHNR). The need for, inverse Fourier transform of the masked cepstrum back into the frequency domain, baseline correction and the usual harmonics-to-noise ratio (HNR) calculation is avoided by this approach. The technique is examined using synthetically generated voice signals.

Full Paper

Bibliographic reference.  Murphy, Peter J. (2000): "A cepstrum-based harmonics-to-noise ratio in voice signals", In ICSLP-2000, vol.4, 672-675.