In attendance: Jason, Alastair, Greg, Hannah, Saining, Brian, Anton, Ich.
Run-throughs for the SIMSSA workshop will happen next Wednesday, but individual presentations should be mostly polished by then; there are many presentations so not much time to spend on each one.
Alastair suggested throwing together a group overview document to deal with higher-level RODAN documentation.
- Anton has been working on issues related to Django and server-side job restarts. He reports that restarts are now handled properly in the system, when restarting from both exceptions and crashes, and has been documenting as he goes. In addition, he has made slides for the SIMSSA presentation and sent them to Andrew for revision. His last day of work (a week from Friday) is the day before the workshop, but he assured Ich that he would be in attendance!
Saining was out sick part of the previous week. She just received new documents from Matt, and will use them to try relation extraction. She reports that extraction thus far has been finicky, as the program is highly context-dependent.
Ich pointed out that even folks in the natural language processing (NLP) field know that the relations extracted have to be fairly basic in order for it to work well. Alastair mentioned that at the CompMusic workshop there was a presentation of work done on extracting relations from forums on Indian Classical music; the paper will be published in September.
- Alastair, back from the CompMusic workshop in Turkey, told us about Istanbul. He spent three days learning three new types of music and then two days learning about computational methods tailored to analyzing them. He went to lots of concerts, ate good food, and even took a boat to Asia for dinner! (Istanbul straddles the continental divide....)
- Hannah has been continuing working on rule application/visualisation for CRIC. She also introduced her thesis topic to the lab: an implementation/comparison of two audio-to-audio alignment algorithms.
- Jason has been working on the automated accompaniment portion of the score tracker for CRIC. His last week's issues with noise boiled down to an "operator error" problem: it turned out he was trying to use higher level features in Antescofo than were necessary. Now it runs better! He is currently trying to determine which kind of moving averages best suit our needs. He also wrote documentation for using Antescofo and Noteability Pro, which is now up on the forum. For his thesis, he has been setting up interviews and helping Echonest implement his downbeat detection model.
- Greg has been fixing bugs from importing NEON into RODAN. He created pluggable style libraries and added the ability to have multiple pieces on one stave. He mentioned that NEON does not assume equidistance between staff lines for every stave on the page(!), only between lines within a single stave. Next up: adding additional neumes. He also introduced his thesis topic to the lab: an online, genetic algorithm-based guitar tab transcription/prediction program called Robotaba. Brian suggested that when he finishes his thesis he should summarize it into a one page dumbed-down version for lazy guitarists.