Designers of VT systems know that you may be able to fool the eye, but you can't trick the ear. Perhaps the most important aspect of a good videoconference is audio quality. If voices come through gargly or haltingly, or if lip-synch is obviously delayed, users will be turned off; you'll have a hard time convincing them that videoconferencing is a good idea.

Studies have shown that 100-ms delays are detectable, but tolerable; 250-ms delays are annoying; and 450-ms or greater delays are unacceptable. Later in this chapter, we'll examine the contributions of network latency and jitter to the overall problem.

Spurred not only by visual telephony, but also by a big market for good quality IP telephony, the industry has settled on the G.7xx audio algorithms shown in the following table. An ideal algorithm consumes the lowest amount of transmission bandwidth while delivering a wide audio frequency spectrum. In VT systems, bandwidth not used by audio can be made available to improve picture quality.

Implementing an ideal audio algorithm requires a more complex codec. In the early days of H.320 system development, when dual 56-Kbps modems provided most videoconferencing connections, G.711 was the only audio algorithm available, and H.261 was the only standard video algorithm. Channel management was simple: Send audio across one channel; send video across the other.

G.711 reconstitutes only a "telephone quality" frequency spectrum. To improve H.320 videoconferencing systems, G.722 was introduced as a "high fidelity" audio option. Then, when inverse multiplexing became available for combining multiple 56-Kbps or 64-Kbps telephone carrier channels into one videoconference transmission band, G.728 offered the option of compressing audio to only 16 Kbps so that remaining bandwidth could be allocated to video.

In the ITU proposals for H.323, the VT architecture for low bandwidth operations over POTS and TCP/IP LANs, a new algorithm was needed to compress audio substantially. G.723 is required for H.324 VT over POTS; it still is considered optional for H.323 LAN use. We recommend that you make G.723 audio a mandatory requirement in LAN/WAN-based systems that you will use for collaboration and training.