27 March 2008

What's Missing From Pandora - Mood

This week I thought I would offer up an unconventional resource to the class - a conversation with an expert.

Last week in class we had a very interesting discussion about the music recommender, Pandora. I was motivated to contact one of my old friends who is an expert in music, radio, and computing.

Me:
"Some members of my class have had some happiness using a music recommender that you might have some fun with; it's called Pandora. It started out life as a project to develop a DNA for music. It is pretty well described at en.wikipedia.org/wiki/Pandora_(music_service) but the stuff I thought you might find interesting is the list of attributes that Pandora employees assign to each piece of music. The list is at en.wikipedia.org/wiki/List_of_Music_Genome_Project_attributes."

Friend, Initial Response:
"Well, I played with Pandora some.

As a former DJ and musician, I think I have a pretty good idea of how this should work.

It seems to match eras more than sounds. It put Tired of Waiting by the Kinks with The Who. Its a VERY different sound and style. I don't get that. Also put 19th nervous breakdown with the Who. No again. It also put Tom Petty in here. Close but not really.

Also put Bos Skaggs with Steely Dan. NO. Steely Dan is jazz. Bos Skaggs has horns but he is not jazz. Reggae?

It did better on Miles Davis. Art Blakely and the Jazz Messengers, etc.

I'll play with it some more. This is a massive undertaking, creating formats on the fly. Certainly interesting."

Friend, Insights Begin:
"I think that I have some insight into what I think is missing from Pandora. They are arranging these songs by objective measurements (male vocal/reggae beat, etc.) but do not have anything subjective in them.

It would be possible for them to group a song with revolutionary content (Street Fighting Man, Volunteers, etc.) with a pacifists song (Bargain, Revolution (Beatles)) based on these measurements.

I think of music in more subjective terms: what feelings, thoughts, and other experiential dimensions does the song evoke? What state changes occur within the listener? What is the objective of listening to this song (calming, introspection, resolution, seduction, whatever.) and the type of artistic treatment. What is the INTENSITY? Joni Mitchell is emotionally INTENSE but she is folk/jazz/rock. This measurement is very important in this category and is probably important in others.

These other attributes may get you in the neighborhood but they don't really achieve the objective for me at least, of meaningful picks. I spent many years selecting music. In broadcasting, you try to achieve a flow or transitions up and back down. Too much of one type fatigues listeners so the music moves up and down in waves.

I designed a computer program to chose music based on some of these criteria once. It was based on what I thought mattered overlaid onto what was being done in radio at top stations at the time. Radio since then has gone backwards instead of advancing, with the exception of what is going on in satellite radio......"

Me:
"Most interesting. One member of our class has used Pandora as part of her class work but, is not very impressed. She said things that showed that she was more concerned about the emotional side of music. She often used the word 'mood' in her descriptions."

Friend, The Magic Word - Mood:
"MOOD is very important in radio.

I was lucky to have gotten to work with one of the best program directors in the history of radio. The whole thing is about mood and subconscious/unconscious messages.

We were sensitive to things you wouldn't believe and have exact phrases that we followed. Some songs are allowed only in the evenings, some are banned from the mornings. It is all about matching and leading moods."

1 comment:

wilson said...

Thanks for posting this interesting conversation. It's fascinating to hear what each individual wants from a radio station.

It's true that the Music Genome Project uses mostly objective measures to categorize songs. (though the list a fan posted on Wikipedia is not complete or accurate, actually...)

We find that one listener would categorize a song as "happy" while another would categorize it as "sad." Same thing with the potential definition of "calm" music. The categorization would be all over the map.

Among the hundreds of traits we do measure, there are a lot of things that help to determine the mood of a song, however. For instance whether it's in a major or minor key, the bpm, dissonance, danceability, and a bunch of tags for lyrics. (political? uplifting? romantic? aggressive? religious?)
We do have several tags for "intensity" as well.

The date the song was published (or 'era') is in there as well, but it's only 1 of hundreds of traits, so it's not a major contributor in matching songs.

Thanks for the interesting discussion!


Lucia, from Pandora
[lucia at pandora dot com]