Difference between revisions of "2008:MIREX2008 Plenary Topics"

From MIREX Wiki
(MIREX 2008 Plenary Topics=)
(Comments)
 
(6 intermediate revisions by 3 users not shown)
Line 1: Line 1:
 
Greetings:
 
Greetings:
 +
 +
The MIREX 2008 plenary meeting will be held Wednesday, 17 September as a working lunch meeting at 13:30-15:00. This will be followed by the MIREX 2008 Poster Session which will be held 16:00-18:00.
 +
 +
Please post any questions or comments you might like the MIREX community to discuss during the meeting.
 +
 +
==Comments==
 +
 +
It would be nice to have a discussion on potential new MIREX tasks: singing voice segments detection, music structure detection
 +
 +
Also a discussion about the possibility to have a new test set for audio chord detection
 +
 +
Qualitative tasks?
 +
 +
Possible changes in QBSH task to make it more real-world compliant (by Roger Jang):
 +
# More noise MIDIs to make the song database size > 10000 ===> Does IMIRSEL has enough computing power?
 +
# Corpous contribution from participants ===> Format? Tools? Guideline for corpus preparation?
 +
# Evaluation criteria that takes care of multiple versions of ground truths (to avoid human selection of noise MIDIs) ===> Top-10 recognition rate?
 +
# Release all corpora after each MIREX
 +
# Timing constraint for each submission ===> 10 seconds for each query?
 +
# Heterogeneous song database ===> MIDI, WAV

Latest revision as of 12:23, 17 September 2008

Greetings:

The MIREX 2008 plenary meeting will be held Wednesday, 17 September as a working lunch meeting at 13:30-15:00. This will be followed by the MIREX 2008 Poster Session which will be held 16:00-18:00.

Please post any questions or comments you might like the MIREX community to discuss during the meeting.

Comments

It would be nice to have a discussion on potential new MIREX tasks: singing voice segments detection, music structure detection

Also a discussion about the possibility to have a new test set for audio chord detection

Qualitative tasks?

Possible changes in QBSH task to make it more real-world compliant (by Roger Jang):

  1. More noise MIDIs to make the song database size > 10000 ===> Does IMIRSEL has enough computing power?
  2. Corpous contribution from participants ===> Format? Tools? Guideline for corpus preparation?
  3. Evaluation criteria that takes care of multiple versions of ground truths (to avoid human selection of noise MIDIs) ===> Top-10 recognition rate?
  4. Release all corpora after each MIREX
  5. Timing constraint for each submission ===> 10 seconds for each query?
  6. Heterogeneous song database ===> MIDI, WAV