The goal of bandwidth extension of speech (BWE) is to extrapolate the missing low or high frequency components of the wide-band speech (50.8000 Hz) based entirely on information contained in a narrow-band signal (300.3400 Hz). In this paper we propose a new method for high-frequency regeneration of the excitation signal, using the correlation between the shape of the glottal flow waveform and the spectrum of the voice source. The high-band excitation is generated by performing a pitch-synchronous timescale (PSTS) transformation on the linear prediction narrow-band residual to generate an high-pass signal that retains the periodic characteristics of the original signal but with a larger open quotient. This method is easy to implement and does not introduce discontinuities in the spectrum of the regenerated excitation. It can be used in applications for BWE where no side information is transmitted or for low bit coding of wide-band speech.
Cite as: Cabral, J.P., Oliveira, L.C. (2005) Pitch-synchronous time-scaling for high-frequency excitation regeneration. Proc. Interspeech 2005, 1513-1516, doi: 10.21437/Interspeech.2005-530
@inproceedings{cabral05b_interspeech,
author={Joao P. Cabral and Luis C. Oliveira},
title={{Pitch-synchronous time-scaling for high-frequency excitation regeneration}},
year=2005,
booktitle={Proc. Interspeech 2005},
pages={1513--1516},
doi={10.21437/Interspeech.2005-530},
issn={2308-457X}
}