Split band CELP (SB-CELP) speech coder

We discuss the split band code-excited linear prediction (SB-CELP) speech coder which employs an iterative version of the harmonic sinusoidal coding algorithm to encode the periodic contents of speech signal. The speech spectrum is split into two frequency regions of harmonic and random components and a reliable fundamental frequency is estimated for the harmonic region using both speech and its linear predictive (LP) residual spectrum. The resulting sinusoidal parameters are interpolated to reconstruct the periodicity in speech waveform. The level of periodicity is controlled by computing a cutoff frequency between the harmonic and random regions of spectrum. The random part of spectrum and unvoiced speech are processed using the CELP coding algorithm. The SB-CELP speech coder which combines the powerful features of the sinusoidal and CELP coding algorithms yields a high quality synthetic speech at 4.05 kb/s