The design of a new codec at 32 kb/s for audio and high quality speech (bandwidth limited to 7 kHz and sampled at 16 kHz with 16 b/sample) is presented in this paper. This codec is a good substitute for the G721 ITU Standard and its 64 kb/s variant G722 that are based o More
The design of a new codec at 32 kb/s for audio and high quality speech (bandwidth limited to 7 kHz and sampled at 16 kHz with 16 b/sample) is presented in this paper. This codec is a good substitute for the G721 ITU Standard and its 64 kb/s variant G722 that are based on ADPCM and dating from the late 1980s. This new codec comprises adaptive wavelet transform coding, psycho-acoustic modeling, quantization and variable length entropy and run-length coding. The novelty here is the use of a parametric wavelet kernel and the way the wavelet packet tree (WPT) is expanded so that better matching is achieved with critical acoustic bands. The explicit kernel permits to control the sharpness of the basic half-band filter of which the filter used in the Fast Wavelet Transform (FWT) coding are derived. The psycho-acoustic modeling of MPEG1-Audio is used but instead of employing power spectrum for calculating the Signal-to-Mask ratio (S/M), we have directly used the energies of WPT output signals. As a consequence, the computation cost is reduced. The number of quantization bits in each band is controlled by the corresponding S/M ratio. The Variable Length Coding (VLC) used here is an extension of JPEG Huffman coding where some modifications are made to adapt this scheme to speech characteristics. The developed codec has the capability of reducing the bit-rate and controlling the required quality by changing the S/M ratios. Therefore, it can be used for fixed capacity channels by the same token. It is shown that this scheme has a very good quality at 32 kb/s and that the coded signal is quite indistinguishable from the PCM signal digitized at 16 kHz and 16 b/sample.
Manuscript profile
