In attendance: Jason, Gabriel, Wendy, Alastair, Hannah, Brian, Catherine, Andrew, Ich.
Pre-meeting discussion
Brian's wrist is worse than previously thought
Individual updates
- Jason recently released a single on vinyl which is now in the top 10 of drum and bass releases in current charts, in terms of sales. Several thousand records were printed, at 12". The original track was written by Jason in collaboration with a student about 2 years ago, and was released with Offshore Records as a CD compilation. This version is a redux, with analog synths to improve the production quality. There is a remix by Deep Blue on the flip side. On the thesis front, Jason has been writing about samplers, particularly voices and polyphony and how they are related, and about the tradeoff between sample rate and memory.
- Gabriel has been working on the algorithms for the bar-finding app. He has discovered that workflows are very image-specific - it's hard to find a particular workflow that will work well for many different images. Gabriel will be leaving for Peru tomorrow, as part of his tour with Lucybell; Lucybell will be playing in Montreal on September 14, the details of which will be emailed around to everyone shortly. Gabriel mentioned that the laptop (Macbook Air) that he currently uses is the one that he uses when playing live, and that the patches for all of his songs are on his github. He has also been working on his thesis - over the last month and a half, he did some reading of music listening behaviour literature, especially in psychology research. Recently, he did some analysis on Last.fm scrobble data for various users including Ich and Brian. He produced graphs that plotted average number of scrobbles against time in a week, averaged over several months. The intent is to eventually understand the listening behaviour of different types of people based on their scrobbles over time.
- Wendy has been away most of last week due to illness. Recently, she got diva.js working in IE8+ and in other Windows browsers. Previously, there was a bug with all browsers on Windows in which the canvas plugin wouldn't load the entire image; this was simply the result of double variable-declaration and has been fixed. Everything works in IE9, and in IE8, canvas is not supported so the canvas plugin is simply disabled. Next, she will be working on diva documentation and whatever other issues need to be cleared up for the September 1 release of Diva 2.0.
- Alastair has been working on his theis. The first of his three algorithm experiments are done, and he is about to start the second set of experiments. He is trying various things to test the robustness of various fingerprinting algorithms, including: speeding songs up by 5%, resulting in tempo and pitch shifts; using shorter input signals; using GSM audio (with an 8 kHz sample rate) to simulate capturing audio with a phone; global volume adjustment; and variable compressors. The speeding songs up was inspired by the anecdote that some radio stations will speed up their songs in order to cram more songs into the available time (and thus have more time for advertisements). Ich says it would be nice to get evidence of this practice, but it is difficult. Alastair also mentioned getting an email from a PR company for Google in Toronto, asking for an interview about his GSoC experience. Also, Github gave us 10 more private repositories for free, bringing us up to 20 private repos in total (about the equivalent of a $50/month plan). Which is awesome. Github is awesome.
- Hannah does not have much to report - she was gone for the last two weeks, first in Vancouver for a week and then back home to Michigan for another week. She has been working on CRIC stuff lately: unit tests, user-friendly documentation, etc. Constantine got her some sample XML files, and all the rules seem good, with only a few things that still need to be tweaked. She has nothing to report on the thesis side.
- Brian was in Toronto for two weeks, for his brother's wedding, and has been working on the bar-finding correction interface for the last few days. Things are going well on that end. You can move, split and drag staves, and the interface preserves ordering as well as system size. So far, Brian has only tested on the single-staff system that he has; he is waiting on more MEI files to test on.
- Catherine, ironically, spent yesterday finding and numbering bars. She first uploaded a score, in PDF format, to Partifi, which is a web app that does basic staff-finding and then cuts and pastes the relevant parts together (so that all the violin parts are together, and all the cello parts are together, etc). However, Partifi is unable to do bar-finding or numbering beyond the first bar in each system, so Catherine had to number the rest manually, which took about two and a half hours. If only there was some sort of web application that could do it for you. She has also been working on Elvis stuff, and has uploaded the metadata for a book, all of the pieces of which have already been tagged. She then manually organised the tags into various umbrella categories (e.g. the "waltz" and "polonaise" tags would both go under the "dance" category). Ich mentioned that this is an example of an emergent ontology, which is a good idea in these situations - instead of trying to categorise everything at the beginning, we can do it afterwards, and so let it emerge naturally. Also, Catherine is having a housewarming party on Saturday - check your email for details.
- Andrew has been working on processing manuscripts owned by the Islamic Studies department so that they can be displayed using diva. This involves renaming 2 terabytes of image data and converting it to JPEG2000 format. He wrote a tornado server to serve each of the 3326 manuscripts - each of which has up to several hundred pages - with diva, and rewrote the Python divaserve module to use memcached (with the cache files being saved to disk or stored in memory if memcached is not available). To get the dimensions for the JPEG2000 images (required for diva to serve the manuscript properly), he is using ExifTool, which quickly inspects the metadata for each image and returns JSON. This works pretty well - much better than tools like ImageMagick, which have to actually open the image - resulting in lower performance - just to get the dimensions. In other news: the backup server arrived, and it needs a new name. Hannah suggested "fakebook". The new name will be set after the hard drive (SSD) arrives.