COMPRESSION Codecs (VoIP)

3.1
Voice compression is broadly classified as waveform, vocoders, and hybrid codecs [Kondoz (1999)]. Waveform-based codecs of G.711, and G.726 encode-decode voice on an actual signal without making an assumption on speech models. They work on any input that is supported by sampling without any significant distortions.
A vocoder is a voice coder that makes use of a vocal tract voice production model. Vocal tract-based compression achieves a better compression ratio than waveform-based codecs. Vocoder-based codecs generate a set of parameters that represent the speech production models. On the receive side, voice is synthetically reproduced based on the parameters. These codecs take very low bandwidth (very high compression) and deliver lower quality. They are of a moderate complexity for processing. Linear predictive coding- 10 (LPC- 10) codec uses 2.4 and 4.8 kbps [Kondoz (1999)], which is much lower than waveform and hybrid codecs. The main benefit of vocoders is a lower bit rate.
Hybrid codecs achieve acceptable compression and quality. Compared with vocoders, hybrid codecs deliver better quality, and a wideband version of hybrid codecs can exceed waveform-based codec quality. Several extended techniques in hybrid codecs are beyond the techniques used in vocoders and waveform-based codecs. Hybrid codecs analyze the signal in frequency and time domain. Frequency-domain compressions are known as sub-band or transform – domain coding. Time – domain compression makes use of short – term redundancies through linear prediction, a long-term excitation rate through pitch analysis, and residual signals. Linear prediction is used in vocoders without several enhancements appearing in hybrid codecs. The waveform codecs of G.711 and G.726 and the hybrid codecs of the G.729 family are the most popular for VoIP applications.
The classification given here does not prevent any reuse of the benefits from one class of codecs to another. Waveform-based G.711 makes use of LPC for voice activity detection (VAD)/comfort noise generation (CNG) algorithms (discussed in topic 4), and pitch and linear prediction in packet loss concealment (PLC) operations (discussed in topic 5). Wideband codec G.722 is typically considered under waveform coding, and it makes use of ADPCM, but G.722 PLC extensions of the codec [ITU-T-G.722 Appendix III (2006)] make use of extended techniques of hybrid codecs. As a conclusive note, waveform and hybrid codecs are the most popular, and vocoders are not used in VoIP. In the naming conversion, hybrid codecs are known by name as low bit rate codecs, even though “low bit rate” does not convey any separate classification.