5 Years

My brain hurt like a warehouse, it had no room to spare
I had to cram so many things to store everything in there

Five years ago, well a couple weeks shy of five years, I began my PhD research under the supervision of Mark Sandler at the Centre for Digital Music (C4DM) in what was then the Department of Electronic Engineering at Queen Mary, University of London. We were a group of 30 or so researchers spread across two offices with a shiny new Listening Room to use in our research.

C4DM in November 2009

C4DM in November 2009

Now we’re something closer to 60 researchers associated with the group spread across four offices (once the construction dust is settled on the new office space). We are also now a part of the School of Electronic Engineering and Computer Science and within qMedia. Along with the Listening Room is a new Control Room and Performance Lab to be used with the Media Arts and Technology doctoral training centre.

Rocking out in the Listening Room

Yves and Matthias in the Listening Room in April 2007. Not many people know this, but Yves is a particularly talented trumpeter.

This September will commemorate 10 years of C4DM as a research group, and it will be the first September since 2005 that I will not be a researcher within it. I submitted my PhD thesis in July 2010 and have been a postdoctoral research assistant since then. It’s been an amazing time, but universities need turnover – they thrive on a continuous flow of new people with new talents and passions – so it is time to move on.

Presentations on MIR to the public at the Dana Centre.

Presentations on music information retrieval to the public at the Dana Centre as part of the OMRAS2 project in 2008. Photo taken by Elaine Chew.

I’m not going very far (in fact I’ll be back tomorrow for a meeting), but it still feels like a significant change. I’m not ready to publicly announce my next steps, but I’m happy to talk with you in person about what exciting things I will be up to past August. I’ll be talking about my new venture at the C4DM 10th Anniversary Event: Past, Present and Future, so come along to that! Even if you’re not bothered about what I’ll be up to, come along anyway! There’s also an evening event.

C4DM stand at the Big Bang in 2010

The C4DM exhibition stand at the Big Bang Science Festival in 2010.

Goodbye, C4DM!

Why MIR Researchers Should Care About User Experience

I’m at NYU for a couple months to conduct a rather large user study on an interface that I’ve developed. I talked about my PhD work and my current work with a gathering of PhD and MS students last week. I’ve included my slides below, though we didn’t really go through them. It’s an updated version of a talk I gave at a couple universities last autumn.

On a related note, if you look around and find that it’s June and that you seem to be in New York, why not contribute 2 hours towards science? You’ll even get a shiny $20 bill in return. Go here for more information.

Some good discussion ensued and it helped me solidify my own thoughts, particularly as to why MIR researchers should care about user experience. To people within HCI or UI/UX or design fields, to not consider the end user seems like a ridiculous notion, but it’s unfortunately an issue that’s only beginning to be addressed.

I talked with a couple researchers that are working on an interface for a library’s music collection. As we discussed ideas and played the game of “oh, have you heard of ______, let me google that and find it for you,” it was admitted that they didn’t actually know why or how users use would the interface they were creating. They didn’t know what are the most common tasks that a patron of the library’s music collection is trying to accomplish. They said the librarians weren’t even sure. One of the researchers I was speaking with said that in his undergrad he spent a lot of time on ethnographic study and user-centered work, but that the work he was doing now was much more focused on the engineering. Unfortunately, I think that means that they’ve developed a piece of software that might not be useful to anyone.

If every researcher has a platform that they like to shout from, these are the main placards I would display around my soapbox:

  • The data should not dictate how people interact with that data, people should dictate how people interact with that data. My pet peeve is that just because you can project a multidimensional timbre space onto two dimensions doesn’t mean you should. At the very least you shouldn’t force a user to interact with it.
  • Don’t kill new products by letting the first release be your first user testing. Genius had negative reactions when it was first released, and I don’t think I’m the only person that abandoned it after some poor recommendations. Google Music is now facing very public discussions of poor performance that perhaps could have been addressed through some private studies before a semi-public release. If something doesn’t work right away for a user, then you may have lost them forever.
  • Users can tell you which 1% problems to focus on. If the end user can’t perceive any difference in the 1% increase in the performance of your algorithm, perhaps it was a wasted effort. There could be a 1% increase in another aspect of the system that would actually improve the system as a whole.

My current work is attempting to address how a small collection of music (e.g. a list of search results returned from a query) can be best presented to a user. I’m doing this without considering a specific MIR backend – I want to separate how the user interacts with the data from how that data was collated. It’s a very specific use case, but because of that I think it’s one that can be easily improved. I’ll certainly be posting more about it as the work is completed.

One in a Million

So I’m designing an experiment. I have an interface. I think it’s really neat, but would like to measure the best way to use this interface and try to quantify its neatness factor. I don’t want to tell you exactly what it is, because you might be a participant in my user study and that could muck around with the results. Suffice it to say that this is an interface for music search and browsing.

When I say it’s an interface for search and browsing, I don’t mean it performs a query. It is only a means to navigate a collection of music. How that collection of music has come into existence is someone else’s business. I just want to help people interact with a collection, in this case smaller collections. The idea is that someone performed some kind of query on the world of music and has a small (< 50 songs) set of songs that they want to traverse through in an efficient manner.

The Problem

I need a several sets of songs in order to perform my user study with my interface. Participants will be searching for specific songs and browsing for songs that fit a written description. As this is an interface for music search and browsing, I think that those sets of songs should be thoughtfully chosen.

I need

  • 6 sets of heterogeneous songs.
  • 10 sets of homogeneous songs so that there are 2 sets of a single “genre” for 5 “genres.”
  • All sets needs to be unique and no song appears in more than 1 set.
  • The sets have no order.
  • There will be approximately 30 songs in a set. This may change slightly after some pilot studies, but it shouldn’t change significantly.

Heterogeneous songs are songs that are as different as possible in timbre and musical style. I want as little similarity between songs within a heterogeneous set as possible.

Homogeneous songs are songs that are as similar to the other songs in the set as possible. This includes notions of “genre” and timbre. I want songs that are similar in signal content and musicality.

I want to use songs that are from the Million Song Dataset. I want this to be reproducible research, and I want to use a dataset that will overlap with other studies. Plus, the 30 second audio clips are exactly the audio content I want for the study – I don’t want full songs.

So I want to know how do I choose my 16 sets of music from the million available songs? I don’t want to write a lot of code – I’m not interested in this selection as a research question. I just want to do it and have it be reasonable. I’d like to use a combination of the Echo Nest API and the sample code for the Millon Song Dataset, but pointers to other useful bits of code will be appreciated as well.

My two main questions are: how should I choose my song sets and what “genres” should be represented?