Posts Tagged playlist
Controlling the artist distribution in playlists
Posted by Paul in code, data, Music, The Echo Nest on January 12, 2012
The Echo Nest engineering team just pushed out a new feature giving you more control over the artist makeup in playlists. There is a new parameter to the playlist/static API called distribution that can be set to wandering or focused. When the distribution is set to wandering the artists will appear with approximately equal distribution in the playlist. If the distribution is set to focused artists that are more similar to the seed artists will appear more frequently. When combined with the variety parameter, you have excellent control over the number and distribution of artists in a playlist. If you want to create a playlist suitable for music discovery, create a playlist with high variety and a wandering distribution. If you want to create a playlist that more closely mimics the radio experience choose a low variety and a focused distribution.
I’ve put together a little demo that lets you create playlists with different levels of variety and distribution settings. The demo will create a playlist given a seed artist and show you the artist distribution for the playlist. Here’s the output of the demo with distribution set to focused:
You can see from the artist histogram that the playlist draws more from artists that are very similar to the seed artist (Weezer). Compare to these results from a wandering playlist with the same seed and variety:
You can see that there is flatter distribution of artists in the playlist. You can use variety and distribution to tailor playlists to the listener. For instance, you can give the Classic Rock Radio experience to a listener by setting variety to relatively low, setting the distribution to focused and seeding with a classic rock artist like Led Zeppelin. Here’s the artist distribution for the resulting playlist:
That looks like the artist rotation for my local classic rock radio.
Give the demo a try to see how you can use variety and distribution to match playlists to your listener’s taste. Then read the playlist API docs to see how to use the API to start incorporating these attributes into your apps.
The Demo: Playlist Distribution Demo (source)
Searching for chilled metal
It seems like every heavy metal band has at least one chill-out song – from Metallica’s Nothing Else Matters to Led Zeppelin’s That’s the Way. These tracks give some relief from the otherwise relentless pounding of the hammer of the gods. It’d be nice to be able to collect up a bunch of these chilled-metal songs into a playlist – perfect for when your mom’s visiting (she tells me that she doesn’t like pounding metal).
ChilLed Zeppelin - Some rights (by-nc-sa) reserved by Heinrich Klaffs
To find chilled metal, we can use The Echo Nest API. The Echo Nest has calculated a wide range of acoustic and musical attributes for millions of songs. One such attribute is energy . We can make a simple song/search query for heavy metal songs that have low energy. These will be our chilled-metal songs that your mom enjoys so much. Here’s the API query:
http://developer.echonest.com/api/v4/song/search?&style=heavy+metal&sort=energy-asc
This query searches for songs by heavy metal artists, and sorts the results in order of ascending energy (so the lowest energy tracks will be returned first). The query does a really good job of finding chilled metal. Here’s a sampling of the results:
Sphinx (The Guardian) by Black Sabbath
Sphinx (The Guardian) by Black Sabbath - energy: 0.0003
Demon Driver reprise by Gillan
Demon Drive by Gillan - energy: 0.010
The Return by Saxon
The Return by Saxon - energy: 0.013
Solitude by Judas Priest
Solitude by Judas Priest - energy: 0.049
Joan of Arc by UFO
Joan of Arc by UFO – Energy: 0.05
Fear by Black Label Society
Fear by Black Label Society – Energy – 0.119
You can also use the Echo Nest Playlist API to generate a chilled metal playlist. Here’s a call to create a playlist of chilled metal in XSPF format.
You can toss this playlist into a player like VLC or Songbird that supports XSPF and start listening to chilled metal right away (30 second samples only) like this:
% curl 'http://developer.echonest.com/api/v4/playlist/static?api_key=N6E4NIOVYMTHNDM8J\ &style=heavy+metal&max_energy=.1\ &type=artist-description&bucket=id:7digital&bucket=tracks&limit=true\ &format=xspf' > chilled-metal.xspf
% open chilled-metal.xspf
There you go, you now have all the tools you need to keep your chilled metal queue filled and fresh, almost everything you need to keep your mom happy.
Thanks much to 7Digital for providing audio clips and album art.
Do you use Smart Playlists?
Don’t count the pre-fab smart playlists that come with iTunes (like 90′s music, Recently Added, My Top Rated, etc.). Once you’ve counted up your playlists, take the poll:
Some preliminary Playlist Survey results
People expect human DJs to make better playlists:
The survey asks people to try to identify the origin of a playlist (human expert, algorithm or random) and also rate each playlist. We can look at the ratings people give to playlists based on what they think the playlist origin is to get an idea of people’s attitudes toward human vs. algorithm creation.
Predicted Origin Rating ---------------- ------ Human expert 3.4 Algorithm 2.7 Random 2.1
We see that people expect humans to create better playlists than algorithms and that algorithms should give better playlists than random numbers. Not a surprising result.
Human DJs don’t necessarily make better playlists:
Now lets look at how people rated playlists based on the actual origin of the playlists:
Actual Origin Rating ------------- ------ Human expert 2.5 Algorithm 2.7 Random 2.6
These results are rather surprising. Algorithmic playlists are rated highest, while human-expert-created playlists are rated lowest, even lower than those created by the random number generator. There are lots of caveats here, I haven’t done any significance tests yet to see if the differences here really matter, the survey size is still rather small, and the survey doesn’t present real-world playlist listening conditions, etc. Nevertheless, the results are intriguing.
I’d like to collect more survey data to flesh out these results. So if you haven’t already, please take the survey:
The Playlist Survey
Thanks!
Music Playlist quiz followup
I had a playlist quiz the other day. To recap, I asked, given a set of 6 songs, find the organizing principal and pick a new good song for the playlist. A few attempted to extend the playlist, but only Adam offered a successful match. Here are the seed songs, but this time I also include the album art – which may help you decide what songs fit and what don’t:
- Made to measure – Umphrey’s McGeez
- Diablo Rojo – Rodrigo Y Gabriella
- Livin’ Thing – Electric Light Orchestra
- Two Step – Dave Matthew’s Band
- Vortex – Burst
- Almost Honest – Megadeth
Adam’s suggestion of XTC’s Wake up fits well:
We’ll call this playlist, the squared circle. There are lots more potential album covers for albums in this genre on Flickr: squaredcircle
Help scientists build better playlists
Luke Barrington, a Music Information Retrieval researcher at UCSD, is trying to improve the state of the art in automatic playlist generation. He’s conducting a survey and he needs your help.
If you are interested in helping out, take the survey.
Here are the details from Luke:
With music similarity sites like Pandora.com or iTunes’ Genius feature that recommends playlists, based on a song that we like, our MIR domain of music similarity and recommendation is finding a mass audience. But are these systems any good? Could we make something better?
This is what I’m trying to figure out and I would like to include your opinion in my analysis.
We are conducting an experiment where you can listen to playlists that are recommended, based on a “seed song”, and evaluate these recommendations. We are comparing different recommendation systems, including Genius, artist similarity and tag-based similarity. Most importantly, we’re are trying to discover the important factors that go into creating and evalutating a playlist.
If you’d like to participate in the experiment by listening to and evaluating some playlists, please go to:
http://theremin.ucsd.edu/playlist/
As an incentive, we’re offering a $20 iTunes gift card to whoever rates the most playlists (but it’s about quality, not quantity!)
To learn more, ask questions or make suggestions, feel free to drop me a line.
Thanks for your help,
Thanks for your help,
Luke Barrington,
Computer Audition Laboratory
U.C. San Diego
van.ucsd.edu
