Do you ever spend money on music? If so, I recommend that you use Bandcamp to buy music directly from artists. On their app, when you buy music (songs, albums, physical copies, merch), the artist gets 85% of the money you spend. If you buy on “Bandcamp Friday”, the first Friday of the month, Bandcamp forgoes their cut, sending 100% of revenue to the artist. The next Bandcamp Friday is September 5th, 2025.

Last year, Suno and Udio released products that generate songs based on user prompts. Both services are based on large language models (LLMs), neural networks that are good at synthesising all kinds of data when trained on large enough datasets. To build these models, Suno and Udio likely would have needed millions of high-quality examples: audio, lyrics and metadata for real music composed by real artists. The launch of these products thus raised red flags for those in the music industry: what data did they use? If it was copyrighted, was this use of the data permissible? And if the models were somehow using bits of copyrighted music to build new pieces of music, was this plagiarism?

A few years ago, my boyfriend Robert proposed to me in the pages of Harper’s Magazine—his words embedded in the cryptic crossword! As an homage to my husband’s proposal, as a thank you to that puzzle’s author (Richard E. Maltby, Jr.), and as a gift to our wedding guests, I wrote a variety cryptic in the style of Harper’s. I emailed a copy to Mr. Maltby Jr., and a week later got the thrill of my life: an invitation to print the puzzle in the pages of Harper’s Magazine!

Another holiday season spent at home, partly with COVID, has meant taking another stab at my Kurodoko hobby project. This time I’ve made a simple web interface to interact with puzzles and solve them! You can try it out here: kurodoko.glitch.me.

I wrote a Kurodoko solving program! I’m rather pleased with it. Kurodoko is a puzzle genre invented by Nikoli.

I’ve just published a new resource: loopextractor, an implementation of an algorithm described in our 2018 ICASSP paper (details here) that uses non-negative Tucker decomposition to model the loops in a song.

In a post this summer, I described Google Street View Movie Maker, a script I wrote to download images from Street View to make short videos of routes. I mentioned that it was inspired by an idea for a music video based for “Don’t Shake” by Hollerado. Well, I finally got around to making it! Here it is:

Last week at ISMIR, we debuted the Unmixer: an interface for extracting and remixing loops.

I finished another hobby project last week: a Google Street View Movie Maker! I have long thought that it would be nifty to be able to preview any route on Google Maps as a video. It would be easy to accomplish by concatenating Street View images; the main challenge is to acquire said images automatically.

In 2011, a team of us at McGill released the SALAMI dataset of structural annotations of lots of music; it was the largest dataset of its kind at the time, and still is. Unfortunately, it has never been easy for other researchers to obtain the audio data: we can provide all the metadata required to identify the tracks, but we don’t own the music so we can’t sell it.

I recently updated a hobby project of mine to Github. The goal was to make an image feed where all the images would have matching edges, but where these edges could evolve over time. I’d still like to tweak it and add new types of designs, but a version of it is finished!

HAMR (Hacking Audio and Music Research) is a hackathon event that has been held many times since 2013, and as a satellite event to every ISMIR conference since 2014. I attended this year’s event held at Deezer and really enjoyed it! To team up with others and try to solve a new research problem ASAP was exhilerating—it was almost like trying to solve a puzzle hunt.

Earlier this month, I was invited to give a talk at the European Music Research Conference. I presented an overview of three recent research projects that all grew, in part, from a shared motivation: to understand musical structure—the way a piece of music is organized—not as a flat, one-dimensional, holistic phenomenon, but as a set of conflicting views of a piece, each view with its own rationale, the various rationales sometimes in conflict and sometimes in harmony each other.

In my last year at AIST, I worked on two projects related to nonnegative factorization, and both have been accepted to ICASSP! Even better, each will be delivered at an oral session.

As of February 1st, I have started a new position at Ircam and CentraleSupélec. I am a post-doctoral researcher on a project called “Dig That Lick”, which will analyze melodic patterns in jazz music on a large scale.

Today was my final day at AIST. I am moving to Paris to start a new post-doc at IRCAM! It’s been a wonderful three years here in Goto’s lab, and I am really grateful to everyone in the lab who helped make my time here so enjoyable.

Last week I attended ISMIR 2017 in Suzhou, China, and once again it was an excellent conference, with a high-quality scientific program (every talk was interesting and well delivered) and top-notch hosting from the National University of Singapore: terrific food and venues, and the chartered shuttles to and from the airport saved lots of people a lot of time.

Our article about the CrossSong puzzle is now print! The article was published in the Journal of New Music Research (JNMR) with Open Access, so everyone can view the full article online. I’m also pleased to announce a major update to the online demo of our system.

Earlier this summer, both articles I worked on were accepted to this year’s ISMIR conference. Last weekend I submitted the camera-ready copies for both of them, which I’m sharing now.

Yesterday, we were pleased to receive the acceptance notification for our submission to the 2017 International Conference on Multimedia and Expo entitled: “Classifying derivative works with search, text, audio and video features.”

Yesterday, we learned that our article entitled “The CrossSong Puzzle: Developing a logic puzzle for musical thinking” was accepted by the Journal of New Music Research (JNMR). (Submitted July 2016, revised January 2017.)

Last year, I helped to advise two colleagues on separate projects, and both were recently published. The projects and first authors are: “Modeling discourse segments in lyrics using repeated patterns”, by Kento Watanabe; and “Probabilistic Transcription of Sung Melody Using a Pitch Dynamics Model” by Luwei Yang.

I’m proud to announce the publication of Mathemusical Conversations: Mathematics and Computation in Music Performance and Composition, a book I edited with Professors Elaine Chew and Gérard Assayag. Music and mathematics have deep historical connections that affect both fields today, and each chapter discusses some aspect of this relationship.

You’ve just remembered a great video you saw on YouTube several weeks ago, and want to share it with your friend. But what was the video again?

Welcome to my rejigged website. After years of writing HTML by hand and hosting at my alma mater, I’ve made the full leap to Jekyll management and GitHub hosting. I used this nifty template from Barry Clark as a basis for the site.