Wideband speech compression using CELP and wavelet transform

The paper presents a new scheme for wideband speech coding in sub-bands, where the wideband speech signal is sampled at 16 kHz and 16 bits/sample, and the subband decomposition is performed using discrete orthogonal wavelets. The wavelet transform uses bandpass filters to split the speech signal into two frequency bands. In order to attain low bit rates, high quality speech and less computational complexity, each subband is then coded independently with improved CELP algorithms for the low frequency band and adaptive differential pulse code modulation (ADPCM) coding for the high frequency band. The overall effect is to produce a wideband speech coder that has almost the same computational complexity as a narrowband CELP coding algorithms and the coding rate at 12.8 kbit/s