This campaign aims to evaluate the performance of source separation algorithms for stereo under-determined mixtures, i.e. two-channel audio signals with three sources or more. Three types of mixtures are considered:

* instantaneous mixtures (static sources mixed using positive gains)
* synthetic convolutive mixtures (static sources mixed using synthetic room impulse responses)
* live recordings (static sources recorded one at a time in a meeting room and subsequently added together)

We invite people in the field of source separation to apply their algorithms to one or more types of test mixtures and submit stereo source estimates, consisting of the contribution of each source on the two mixture channels. All algorithms are welcome, including algorithms originally developed for single-channel mixtures or algorithms requiring prior information estimated by listening or visualization (number of sources, source positions, speech vs. music). We also provide development data including reference source signals for training purposes.

http://sassec.gforge.inria.fr/

Pascal Speech Separation Challenge Part II

The objective of this challenge is to separate the speech of two talkers simultaneously reading sentences from the Wall Street Journal (WSJ) speech corpus. The recordings are made using 16 microphones arranged in two eight element circular arrays placed on a table in the center of a reverberant meeting room. The speakers read the sentences from a variety of locations around the meeting room. The separation techniques will be evaluated in terms of the word error rate obtained from a WSJ recogniser recognising the separated signals.

http://homepages.inf.ed.ac.uk/mlincol1/SSC2/

Speech separation challenge

The task is to recognise speech from a target talker in the presence of stationary noise or other speech. You will be provided with lots of training data. Only one signal per mixture is provided (i.e. the task is "single microphone"). Sentences have a simple structure to ensure the overhead of building an automatic speech recognition system is kept small. For full details see

www.dcs.shef.ac.uk/~martin/SpeechSeparationChallenge.htm

Deadlines

* April 7th 2006: submit your 4-page Interspeech paper for reviewing in the normal way
* June 9th 2006: notification of acceptance

Results will be presented at a special session of Interspeech 2006, Pittsburgh (USA) 17-21 September, 2006.

Organisers

Martin Cooke (University of Sheffield, UK)
Te-Won Lee (UCSD, USA)