A complete audio solution for videoconferencing applications is now available from Signals and Software Limited for the TI TMS320C5x family of DSPs.
Consisting of the three standard International Telecommunications Union (ITU) speech coder algorithms (G.711, G.722, and G.728), plus a full-duplex wide-band acoustic echo canceller, the solution offers videoconferencing equipment manufacturers a cost-effective route to solving their audio needs.
G.711 was the first speech coder to be standardized by the ITU to compress the narrowband (300 Hz-3.4 kHz) telephone signal down to 64 kbps, although it is often now used in 56-kbps and 48-kbps derivatives. G.711 is the mandatory minimum standard for all ISTN terminal equipment.
G.722 makes better use of the available bandwidth by compressing a wideband (50 Hz-7 KHz) audio signal down to the same 48-, 56-, or 64-kbps data rate using a subband adaptive differential pulse code modulation (ADPCM) algorithm. Because of the wider bandwidth, the resultant audio quality is much more natural.
G.728, the lastest ITU speech coding standard, offers toll-quality speech at 16 kbps. Of the three speech-coding algorithms, G.728 is by far the most complex using a backward adaptive variety of the codebook excited linear predictive (CELP) technique requiring 37.5 MIPS for full-duplex operation.
The conventional solution to handle echo caused by loudspeaking audio has been the use of voice switches, which only provide half-duplex communication. The advantage of an acoustic echo canceller is that it enables full-duplex communications by the use of adaptive filtering techniques. The adaptive filter models the impulse response of the loudspeaker-room-microphone system and subtracts the estimate of the echo from the microphone path.
Signals and Software Limited has developed a 24-subband system which runs on a single 28.5-MIPS 'C53 with some additional external data RAM. By tailoring the adaptive filter lengths in the individual subbands to suit typical conference environments, the 4K words of on-chip RAM allows echo spans to be covered up to 200 ms. The 24-band structure with a separate adaptive filter in each subband, allows the processing equipment to be reduced enough such that the echo canceller will fit on a single DSP. Also, by splitting the frequency spectrum into subbands, the correlation of speech signals is reduced and the adaptive filters train more readily.