Abstract:
Most voice and audio codecs are optimized for fixed bandwidth streaming channels. There is a growing need for codecs which operate efficiently over variable bandwidth packetized channels, such as Internet telephony. A novel codec structure is proposed which is optimized for packetized channels. It allows dynamic trade-off between bandwidth, quality, and computational load. Block normalization is performed by segmenting the signal into short time epochs of similar amplitude, and renormalizing each epoch, which greatly decreases dynamic range requirements for subsequent coding. The degree of ``intelligence'' used to find these segment boundaries can be tailored to balance bandwidth against computational load. Channel squelch is performed by suppressing transmission of packets with low amplitude. In this way, silent epochs consume zero bandwidth. The threshold for suppression can be tailored to balance quality against bandwidth. Experiments were conducted to evaluate several versions of the codec, examining bandwidth, computation, and quality (as measured by DMOS testing). Results show good quality at bandwidths and computational loads appropriate for home computers with modem connections to the Internet.