In the comments section at the bottom of this page,
you are welcome to indicate your ideas about a new task for SiSEC 2018.
Please mention:
- A tentative title and acronym for this task
- A short description
- Mention the available databases and objectives
- Indicate the approximate expected participation.
In all cases, note that:
- This new task should be related to source separation and/or latent variable analysis.
- By proposing a task, you of course will be given the opportunity to be the corresponding task leader, which means having the full control of how the task will be managed. The SiSEC committee will assist you in terms of hosting results and announcements.
We cannot wait to welcome you in the organization committee !!
1 comment
Title: Orchestral music source separation
Dataset: PHENICX-Anechoic dataset (https://www.upf.edu/web/mtg/phenicx-anechoic) based on Aalto anechoic recordings, the associated score annotations, and the Roomsim simulated multi-microphone recordings
Description: Orchestral music makes a difficult scenario for source separation, characterized by large reverberation, a complex auditory scene comprising a high number of sources of similar timbres and a high number of instruments per source. I propose evaluating the multi-microphone case which is described in this paper: https://www.hindawi.com/journals/jece/2016/8363507/. Because we have annotated scores and automatically aligned scores, we can decide on allowing score-informed approaches. Supervised approaches can use samples from RWC instrument samples database or NSYNTH by Google.
Participation: Most of the NMF approaches based on the (multi)source-filter model or harmonic model can evaluated in this context.