February 2003 - Nr. 2
Can’t Find that Tune? Hum It
TWIG - Software experts have already revolutionized the music business with the MP3 digital music format, which lets consumers download their favorite tunes from the Internet and record them on a blank CD. Now they are working on the next innovation for audiophiles - a form of software that identifies a song by title and composer from a few bars hummed into a microphone. The system will also help musicians determine the tone structure of a tune, according to a spokesman for Germany’s Fraunhofer Gesellschaft, an umbrella organization for some 50 research institutes and working groups. Fraunhofer scientists are making the tradeshow rounds, and most recently showed off their development during Midem, the international music trade fair that draws recording artists, producers and executives to Cannes, France.
Computer scientists around the globe are working on their own versions of the "Query by Humming" melody recognition system. They all face the problem of redesigning text-based databases and search engines for audio content.
On the job in Germany are scientists with the Working Group for Electronic Media Technology (AEMT) of the Fraunhofer Institute for Integrated Circuits. Part of their work is building a database of songs that includes for each piece a set of "metadata" on the composer, artist, category, rhythm, beat and tempo. The Fraunhofer scientists are working on a method by which this information is extracted automatically from the audio data and attached in the form of an explanatory note. "In order to find the requested song, the sound waves generated by the tune being sung are resynthesized by the computer into a written sequence of notes," AEMT scientist Frank Klefenz explained to Fraunhofer magazine. "The input pitch and timing information is essentially translated back into notes." The system then selects the matching piece of music from the database.
The work is a natural progression from MP3 technology, which opened up access via the Internet to a nearly infinite range of recordings. Now what is needed is an easier and faster way to identify and select the desired music. Query by Humming is not the only innovation in the works. Fraunhofer’s "intelligent stereo" project aims to allow users simply to state the title of the desired piece and have the stereo play it automatically; no more shuffling through CDs or music files. And in a twist on Query by Humming, researchers at another Fraunhofer installation, the Institute for Integrated Circuits IIS in Erlangen, have come up with AudioID, in which the system "listens" to a piece, registers features of the song and can then identify it and even distinguish between different versions. The technology has applications in music sales, broadcast monitoring and protection of copyright.
