Mixed wideband speech and music coding using a speech/music discriminator

In multimedia applications such as videoconferencing, users are demanding higher quality speech/audio transmission than the POTS can offer. 7kHz wideband speech/audio offers a good compromise between bandwidth and sound quality. It improves the intelligibility and naturalness of speech and adds a feeling of transparent communication. Currently the only existing international standard for coding such signals is the G.722 wideband speech/audio coder. While its coding quality is satisfactory it leaves much to be desired with its bit rate. The CELP-based approach has been very successful in telephone bandwidth speech coding, but is not suitable for coding non-speech signals because of the assumed signal production model. This paper proposes an alternative approach to mixed speech/music coding, which uses a discriminator to separate music signals from speech, and codes them with the G.722 coder and a G.723.1-based speech coder, respectively. Simulations shows very promising results.