[Elsnet-list] Announcement : Speech Separation Challenge Part 2.
Eval data now available.
mlincol1 at inf.ed.ac.uk
Wed Mar 14 17:27:58 CET 2007
Further to my previous announcement concerning the Speech Separation
Challenge Part 2, please note that the final evaluation data set is now
available from the challenge web site. For full details of the
challenge see below.
Please note the extended deadline for paper submissions of May 1st.
Speech Separation Challenge 2!
As a follow on to the highly successful PASCAL Speech Separation Challenge
(http://www.dcs.shef.ac.uk/~martin/SpeechSeparationChallenge.htm) we are
pleased to announce the Speech Separation Challenge Part 2! The new task
addresses the problem of separating and recognising overlapping speech from
multi-channel recordings made in realistic environments. Results will be
presented at a special session of MLMI07
(http://www.mlmi07.org/index.php?id=home) taking place in Brno in the Czech
Republic from 28-30 June 2007.
The task is to separate the speech of 2 overlapping speakers recorded in a real
meeting room using two eight element microphone arrays, and subsequently
recognise each speaker. Large quantities of development data are provided. The
speakers read Wall Street Journal sentences, and a standard WSJ recogniser is
provided for recognition of the separated speech, although participants are
welcome to use their own recognition systems if they wish.
Contributions are welcomed from a variety of different approaches including
well-known existing algorithms such as blind source separation and beamforming,
as well as novel techniques.
Full details of the challenge plus downloads for the data and default
recognition system are available from the challenge website :
* Submission deadline for MLMI07 papers for the SSC2 special session is
May 1st 2007 (note, this is later than the standard MLMI paper
Centre for Speech Technology Research
University of Edinburgh
More information about the Elsnet-list