Information on Audio Codecs

Codecs are used to convert an analog voice signal to digitally encoded version. Codecs vary in the sound quality, the bandwidth required, the computational requirements, etc.

Each service, program, phone, gateway, etc typically supports several different codecs, and when talking to each other, negotiate which codec they will use.

X-lite has various audio codecs implemented. The implemented audio codecs of the X-lite 3.0 version are DVI4, G.711 uLaw/aLaw, GSM, iLBC as (narrow band audio codecs) and BV-32, BV FEC, DVI4, L16 PCM (wide band audio codecs). Eyebeam version 1.5 has a wider range of codecs available.

G.711A & G.711U

G.711 is a codec primarily used in telephony, released in 1972. It samples the analogue voice input signal using pulse-code modulation (PCM) at a sampling rate of 8000 samples/second.

The G.711 standard has two algorithms: the A-law (mostly used in Europe) and the U-law algorithm (used in North America and Japan).

The G.711 codec encodes the pulse-code modulation samples (sampling rate of 8000 samples/second) to logarithmic 8-bit samples. This will lead to a (8000 times 8-bit=) 64 kbit/s bitstream. Note that this is the datastream without information send in the headers of the packages, so the actual datastream will be higher. Because the sampling rate of 8000 samples/second and the encoding to 8-bit samples are intrinsic properties of the codec, the bitstream can not exceed 64 kbit/s. If one wants to increase the quality of the voice by increasing the used bandwidth (> 64 kbit/s) one has to use another codec.

Various codecs compared

The following sites were used to build the following comparison:

• VoIP-wiki.de (german website)
• Speex.org
• VoIP-info.org - Codecs-page.

Codec	Compression method	Transfer rate	Frame size
G.711	Pulse Code Modulation (PCM)	64 Kbps	20 ms
G.726	Adaptive Differential Pulse Code Modulation (ADPCM)	16-40 Kbps
G.728	Low Delay Code Excited Linear Prediction (LD-CELP)	16 Kbps	0,625ms
G.729/G.729A	Conjugate Structure Algebraic Code Excited Linear Prediction (CD-ACELP)	8 Kbps	10 ms
G.723.1	Mulptiple Maximum Likehood Quantization (MPMLQ)	6,3 Kbit/s	37,5 ms
G.723	ACELP	5,3 Kbps
GSM	Global System for Mobile communications	13 Kbps	20ms
Speex Narrowband (8kHz)	Patent-free audio compression	2.15-24.6 Kbps	20ms
Speex Wideband (16kHz)	Patent-free audio compression	4-44.2 Kbps	20ms
DVI4	ADPCM	32 Kbps
iLBC	Internet Low Bandwidth Codec (free)	13,3 Kbps	30ms
iLBC	Internet Low Bandwidth Codec (free)	15,2 Kbps	20ms
L16 PCM	PCM	128 Kbps
EVCR	Enhanced Variable Rate Codec	8 Kbps

From the maximum voice quality point of view, when enough bandwidth is available, the best quality can be obtained by using the L16 PCM codec. This L16 PCM codec is the same as the G.711 codec, however the L16 PCM codec samples to 16 bits samples instead of the 8 bits samples in the case of the G.711 codec.

Asterisk will resample all codecs internally to 8000 samples/sec. This means using for example the Speex Wide-band (16 kHz sample-rate) or even Ultra-wide-band (32 kHz sample-rate) with Asterisk is useless.

Codecs with a frame-size other than 20 ms, appear to work bad with Asterisk. This may have to do with the IAX-protocol.