A long term harmonic plus noise model for narrow-band speech coding at very low bit-rates

Abstract

This paper presents a very low bit-rate speech codec based on the long-term Harmonic plus Noise Model (LT-HNM). The HNM is known to be efficient in terms of speech signal representation, thanks to the use of natural parameters: fundamental and voicing cut-off frequencies, harmonics and noise frequencies. Besides, the long-term modeling is particularly efficient in reducing the data size of the model parameters. In this paper we combine both approaches, long-term modeling and HNM, to develop a very low bit-rate coder for narrowband speech. The obtained bit-rates are as low as 2.3 kbps with objective listening quality (perceptual evaluation of speech quality PESQ) of 2.3.