G.729 Audio Coder
for the TMS320C5x
by Signals and Software Limited

Software Overview

DSP software which implements the ITU G.729 speech-coding specification. This codes narrowband 
speech (300–3400 Hz, sampled at 8 kHz) to a bit rate of 8 kb/s. G.729, which offers toll-quality 
speech, is primarily intended for use in Digital Circuit Multiplication Equipment (DCME) type 
applications giving an eight-to-one bandwidth saving over standard 64-kb/s G.711 PCM. Other 
possible applications for G.729 include use in the simultaneous voice data (SVD) standard being 
developed within study group 14 of the ITU and 2 B (128 kbs) channel desktop video conferencing.

Features and Benefits

The G.729 specification uses a Conjugate-Structure Algebraic-Code-Excited Linear Prediction 
(CS-ACELP) algorithm. At the encoder, the speech signal is analyzed at a 10ms frame rate to extract 
the CELP model parameters, which are then coded and transmitted. At the decoder the received 
parameters are decoded and used to reconstruct the speech, which is further enhanced by the use of 
a postfilter. Error concealment routines are built into the decoder. The performance of the 
algorithm may be summarized as follows:
Bit rate				8 kb/s
Audio quality
•	Single transcode		•	Comparable to 64-kb/s PCM (G.711), identical to G.726 at 32 kb/s
•	Multiple transcodes	•	Degradation only apparent after four tandem transcodes
•	Errored transmission	•	Better or equivalent to G.726 at 32 kb/s
•	Transcoding delay		•	35 ms
•	Music coding		•	Limited detrimental effects
Transmission properties
•	DTMF/network signalling tones		•	Equivalent to 64-kb/s PCM
•	Modem signals

The software passes the full set of ITU G.729 test vectors.

Processor and System Requirements

G.729
Program Memory (Words)
Data Memory
(Words)
Processing Load (MIPS)*
Code
Data Tables
Permanent
Temporary
Encoder + decoder + postfilter
9.8 k
2.8 k
2.0 k
1.2 k
34.0
Encoder only
7.0 k
2.8 k
1.2 k
1.2 k
29.6
Decoder + postfilter
2.8 k
2.0 k
0.8 k
0.7 k
4.4

Usage Limitations or Performance Considerations

The DSP code consists of five subroutines: an initialization routine and a processing routine for 
each of the encoder and decoder (including postfilter) functions, and an additional routine to 
implement the synchronization facility defined in the specification (this will usually only be 
required to verify that the code passes the test vectors). The initialization routines are normally 
called only on DSP reset. The audio input and output format is 8-kHz linear samples, and the 
processing routines are called once per 80 sample (10 ms) frame. The encoder converts each input 
frame into an 80-bit parameter stream, which is used by the decoder to reconstruct the signal.
Applications of the G.729 algorithm are
•	Speech compression for DCME equipment, thus allowing eight channels to be transmitted over a 
single 64-kb/s link with a 15-ms algorithmic delay (35 ms total including double buffering).
•	Coding of the audio channel for video conferencing and desktop terminals where transmission 
bandwidth is limited, thus providing additional bandwidth for the transmission of video 
information.
•	Speech storage.
•	Simultaneous Voice and Data (SVD) applications for use over PSTN/ISDN.
•	Mobile communications.
•	Private networks.

Availability

Now, under licence, for a one-off payment and/or royalties depending on the commercial application. 
An Application Note is available. Support consultancy for code integration is also available.

Company Background and Contact Information

Signals and Software Limited (SASL), based in Harrow, Middlesex, UK is a design consultancy 
specializing in the area of Digital Signal Processing (DSP). From concept and algorithm design 
through to real-time DSP implementation, SASL is able to offer its clients fast and cost-effective 
solutions to their DSP needs.

Contact:	David Morley
		3 Jardine House
		Bessborough Road
		Harrovian Business Village
		Harrow, Middlesex, HA1 3EX
		United Kingdom
		+44 (0) 181 426 9533
		Fax: +44 (0) 181 869 1182
		e-mail: davem@sasl.demon.co.uk