Dynamic Programming Search of Articulatory Codebooks

One interesting approach to low-bit rate coding of speech employs physiological models of the glottis (the voice source) and the vocal tact (mouth and nasal cavities) [1]. 

Due to the fact that during natural speech production articulatory gestures are formed roughly at a rate of one every 100 ms, there is hope that ultimately speech coders using speech production models can produce synthetic speech of good to excellent quality at a bit rate below that of high-quality waveform coders.