next up previous contents
Next: PhoneJACK/LineJACK cards from Quicknet Up: Voice-over-IP Mini-HOWTO Previous: Pointers to more information

Hardware-based voice compression

Even with a fast modern PC, you do not want to load the main CPU with the task of compressing a stream of digital audio. The human ear is quite sensitive to the small delays of the sound, and the perceived quality of the audio connection decreases drastically if only a few tens of milliseconds of delay creep in along the way. Linux is a multi-tasking multi-process multi-user system, and there is no guaranteed time of response to any given task. One can play with that by raising priority of some processes, or make the entire kernel time sensitive (as does RT-Linux), but since a dedicated dsp can handle the compression with ease, and the hardware is not prohibitively expensive, it seemed a natural choice for this project.


Ed Sternin