Home | Modem Data Pumps | Telephony | Audio | Utilities | Consultancy | Contact Us

G.729AB Speech Coder Software for Texas TMS320C54x/C55x DSPs

This is an implementation of the ITU G.729A(B) voice compression algorithm. Using the Conjugate Structure Algebraic Codebook Linear Prediction Coding algorithm, 16-bit samples of speech made at 8kHz are compressed into an 8kb/s stream. Data is communicated in frame sizes of 10ms. The overall delay of the algorithm is 15ms.

G.729A is a reduced-complexity version of G.729. In G.729AB, the "B" variant involves a voice activity detection function with comfort noise generation to reduce the average bit rate during silence periods. This is especially useful in packet-based systems such as voice over IP.

The Algotron implementation is available as both G.729A and G.729AB, with encoder and decoder as separate items. It is operated with one routine for initialisation and one main routine for each frame, with separate calls for encoder and decoder. The code has been written to allow multi-channel operation by being fully re-entrant, and is also C-callable.

Performance:

In terms of voice quality, the G.729AB vocoder achieves an MOS figure of 3.9, assuming clean input speech. This is slightly better than G.726 ADPCM coding at 32kb/s. With background noise, G.729AB has a slightly lower performance than G.726.

DSP Load:

Availability: NOW, with implementer's guide - integration support offered