A new approach to low bit rate audio coding using a combined harmonic-multiband-wavelet representation

In this paper, we present updated results for the combined harmonic-wavelet based audio coder. This coder first approximates the audio frame as a sum of several sinusoids using a harmonic analysis-synthesis scheme. The reconstructed harmonic signal is then subtracted from the original to give a residual, which is analysed using a wavelet filtering scheme. After each step (harmonic analysis and wavelet filtering), the perceptually relevant parameters are quantised and encoded. The total least squares (TLS)-Prony algorithm is used for the harmonic analysis scheme, and cascaded M-band wavelet transforms are used for analysing the residual. At encoding bit rates of 60-70 kbit/s, this audio coder is capable of delivering better results than the MPEG layer II audio coder operating at the much higher bit rate of 128 kbit/s.