beaTunes News

Saturday, August 18, 2018

You Are Giants!

Back in November 2017 I had asked you to help with tapping to EDM. Many of you followed the call and earned beaTunes licenses (please don't tap anymore, data collection is over). Thanks again for that!! Since then I have analyzed the submitted data and created new annotations for the GiantSteps tempo dataset originally created by Knees et al.

Long story short:

  • It appears that a substantial part of the original annotations were wrong (almost 9%).
  • All user submitted data incl. new derived tempo annotations can be found here (under "Datasets").
  • The new annotations will be used as one ground-truth for this year's MIREX Audio Tempo Estimation task.

The full paper (link below) describing the collected data will officially be published at this year's ISMIR in Paris—if you're attending, please come and chat with me!

Hendrik Schreiber, Meinard Müller. A Crowdsourced Experiment for Tempo Estimation of Electronic Dance Music. In Proceedings of the 19th International Society for Music Information Retrieval Conference (ISMIR), Paris, France, Sept. 2018.

Labels: , ,

Wednesday, January 31, 2018

beaTunes for taps: offer ends today!

Last chance! beaTunes for taps will end today!

If you're interested in a reward license, start tapping.

Labels: , ,

Thursday, January 25, 2018

beaTunes for Taps: Offer ends soon!

Just to give you an update: On average every segment has been tapped more than 3.5 times and all of them at least twice. This is not quite the amount of data I'd like to have, but it's nevertheless soon time to stop collecting and start with the evaluation.

So, if you're interested in earning a beaTunes license, start tapping before 1/31/2018, as that's the official end date for the offer.

Cheers!

Labels: , ,

Wednesday, December 20, 2017

Tapping: Intermediate Results 2

And the gigantic tapping continues! On average, each of the more than 4,000 track segments has now been tapped twice! And almost all of them have been tapped at least once. That means we're close to 100% coverage, with many segments being backed by multiple people.

As you know, this still is not quite enough data, but we're getting there. So please keep at it! Reward licenses are still offered for every 110 segments that have been tapped. Here are the details.

Cheers!

Labels:

Monday, December 4, 2017

Tapping: Intermediate Results

I must say: I'm pretty psyched! On average, every single one of the more than 4,000 track segments has already been tapped once! This is awesome!

Obviously, it's not enough data, but certainly a super solid start. So please keep at it! Reward licenses are still offered for every 110 segments that have been tapped. Here are the details.

Thanks so much for your help!

Labels:

Thursday, November 30, 2017

Tap like a Giant!

Sorry to bother you with this again... but I really need your help. Even, if you just tap along to a couple of tracks. It's much appreciated. Here are the details.

Labels: ,

Monday, November 27, 2017

Enjoy EDM? Want a beaTunes license?

TLDR: Please help me by tapping along to some EDM tracks on this site.

Part of developing beaTunes is actively participating in the MIR research community. This includes working with scientific datasets for training, testing and benchmarking of algorithms. One of these datasets is the GiantSteps Tempo dataset, created by Knees et al. It contains excerpts of electronic dance music (EDM) annotated with a beats per minute (BPM) value.

GiantSteps Tempo is currently one of the very few EDM datasets that are freely available. But unfortunately it has never been properly verified. This means some reference values may be wrong or ambiguous. I'd like to change that.

Verifying the BPM value for a given song is pretty simple: Just tap along and record the taps. Unfortunately, there are some caveats:

  1. Perception: Tempo is not perceived the same by everybody. Some people may say a given song has 70 BPM, others may say it's 140 BPM.
  2. Tempo Changes: Sections of an excerpt are clearly slow, while other sections of the same excerpt are fast.
  3. Size: There are 664 tracks in the dataset.

What does this mean for a thorough verification?

Science Needs Your Help

In essence, multiple annotations for multiple sections for each track are needed. Assuming we need 5 annotations per section and divide each track into 7 sections, we require 664 x 5 x 7 = 23,240 datapoints. I'm happy to tap along to a couple of hundred songs, but tapping to ten thousands of tracks is an impossible task. Left aside the fact that it would only provide a single point of view and therefore not address the perception issue at all.

Why am I telling you this? Because obviously I need your help. While tapping 23,240 samples is very hard for a single human being without going insane, it's actually relatively easy for a large group. If a couple of hundred people tapped just 100 songs each, the task could be solved in about an hour. Plus the collected data would cover many different ways to perceive tempo.

Earn a beaTunes License

So please help me out! To make it worth your time, I'm more than happy to throw in some beaTunes licenses. To be exact:

Anybody who taps 110 songs and does so well (submissions are checked) gets a free beaTunes 5 license.

To start tapping, please visit this site. Even if you tap just a couple of songs, your submission is much appreciated! And be assured, results will eventually be published.

Thanks so much!

P.S.: If you don't feel like tapping yourself or simply hate EDM—perhaps you know someone who'd enjoy this little task. Please share this.

Update

This offer will end on 1/31/2018.

Labels: , ,