New gig at Polyphonic HMI
I’ve been pretty silent for well over a month so wanted to give an explanation.
Six weeks ago, after nearly six years at Live365, I started a new, part-time consulting gig with a Barcelona-based company called Polyphonic HMI (I’m also planning a new venture and will be seeking funding in September). As you may recognize from earlier posts, I’m a big believer in social-based music and media discovery; in many ways this best reflects the way that people learn of new music, film, books, etc today, in the offline world.
However, I’m also intrigued by the potential of pure technology-driven discovery, which itself falls into one of two camps: (1) analysis of the underlying attributes of music (a la Pandora), and (2) comparisons of consumption patterns among a pool of media/music consumers, aka collaborative filtering (a la Last.fm). Polyphonic HMI (“PHMI”) takes a cue from the former but accomplishes its analysis NOT through the use of humans to subjectively classify a track along X number of attributes but rather by identifying these attributes mathematically, objectively.
By identifying meaningful but sometimes subtle attributes of a piece of music, PHMI can then make direct comparisons with other works. If a label is considering signing a new artist, or determining the first record for push to radio, use of PHMI’s technology can ascertain whether a song from that artist follows mathematical patterns that are similar to songs that sold well in the past, by format/genre or in general. Similarly, the technology can compare a song that I like to myriad others that I’ve never heard but are likely to like.
Unlike the application of collaborative filtering to music consumption patterns (as with Last.fm), there is no “cold start” issue in providing recommendations, i.e. an inability to get new content into the system for recommendations until a sufficient number of people have expressed preference for that content. And unlike the evaluation of attributes by people (as with Pandora), there is no scalability issue; the existing (and future refinements of the) PHMI technology can be readily applied to thousands or even millions of songs to derive a meaningful result for the evaluation of sonic similarities.
As you might expect, analysis with the first objective (commercial success) is being targeted at record labels or other aggregators primarily. Analysis with the second objective (song-to-song or personalized recommendations) is aimed at publishers and digital music providers.
PHMI’s technology was highlighted in a recent issue of the Economist (subscription required), and if you’d like to hear more about the technology or even see a demo, feel free to contact me (email@example.com).
Filed under: discovery, ventures | Leave a Comment