CELT: A Low-latency, High-quality Audio Codec
(with Jean-Marc Valin)
Ever wondered why your telephone doesn't sound as good as your music player? Historically, low-latency speech codecs such as Speex have had poor performance on music or other general audio, since the techniques they use do not scale well to high sampling rates. High quality general purpose codecs such as Vorbis have too high a latency to be used for telephony. We present a new codec, CELT, which has both extremely low latency and good quality at high sampling rates. This makes it possible to play live music together with someone over a DSL connection or enjoy "CD-quality" video conferencing, and enables or enhances a host of other interactive applications. Unlike mainstream audio and video, there is no entrenched proprietary codec in this domain, and our open source alternative is already better than the competition.
This talk will explain why latency is an issue in interactive applications, briefly describe how we were able to extend the techniques used in speech codecs beyond the sampling rates where they are traditionally effective, and talk about the libcelt API and how to write low-latency audio applications for Linux. It will include a live demo of the codec.
Jean-Marc Valin has a B.S., M.S., and PhD in Electrical Engineering from the University of Sherbrooke. He is the primary author of the Speex codec, which provides a free alternative to patented, proprietary speech codecs. He joined the Xiph.Org Foundation in 2002, just after Speex was created. From 2005 to 2008, he was a post-doc at the CSIRO, where he started working on the next-generation audio codec named CELT. His expertise includes speech and audio coding, speech recognition, echo cancellation, and other audio-related topics.
Timothy B. Terriberry received dual B.S. and M.S. degrees in both Mathematics and Computer Science from Virginia Tech in 1999 and 2001, respectively, and a Ph.D. in Computer Science from the University of North Carolina at Chapel Hill in 2006. Since 2002 he has been a volunteer for the Xiph.org Foundation, a non-profit organization that develops free, open multimedia protocols and software. He is the primary author of the Theora specification, and contributed to several of the mathematical components of the CELT codec, enabling optimal encoding of quantized data. His research interests include audio and video compression, motion tracking, target recognition, medical image analysis, computer vision, optical character recognition, and general purpose computation on GPUs.