Musical Detective Work

CSI - New Orleans: Was that really Bix Beiderbecke on "Baby Won't you Please Come Home?" Without a Trace: And where did you hide the Nine Inch Nails? One new computer technology helps ID musicians by their sonic fingerprints, while another can listen to songs and categorize their genre and emotional content.

Two music professors are literally turning off the sound to help identify performers. When Gary Westbrook and Tom Smith try to figure out who REALLY is playing the cornet on "Baby Won't you Please Come Home," they don't trust their own ears, and their musician's sense of each player's unique style. It's their computer that is doing the analysis, and quietly, according to the Chronicle oh Higher Education.

Westbrook and Smith are using sound wave analysis software (SpectraPlus) to study the frequency spectrums of various musicians in hopes of helping identify some unknown works. They believe that each player has a unique spectral signature that runs through all their playing. Although some experts disagree -- citing variations in microphone placement, background noise, and the sound of the rest of the band -- the pair thinks they know which songs actually contain Bix, and which contain his frequent substitute Andy Secrest.

It has long been known that the printed accounts of who played what on many early jazz recordings are unreliable. Sometimes players used assumed names for some sessions, while other times they missed their appointments and had others cover for them. In particular, Secrest was known as a skilled mimic of Beiderbecke, an alcoholic who was often in poor physical condition at recording time.

Combining technical analysis with biographical research, the duo hopes to put forth their hypotheses in a book called "The Jazz Detectives." Jazz aficionados will be unswayed, perhaps, by an argument about which player has the "highest 11,000 hertz level," but some mysteries exist for the debating, not the solving.

Ogihara envisions filters that can help identify songs that you want to hear, and deliver them at the right time for you to hear them. His software uses wavelets to analyze signals and patterns in the songs, and then decide which categories it belongs to. By learning a listener's tastes (and preferences for what kind of music matches what location, or time of day), a computer could then pull out the appropriate music for the occasion (serve up happy jazz when I'm down, syrupy country when I'm up, and skip the gansta' rap when Momma's around).

Other users see applications in helping people label and organize their growing collections of music. In one newspaper account, a 21 year old college student with thousands of songs on his Apple iPod bemoans how long it takes to put in ID tags for 5 gigabytes of music, and expresses interest in Ogihara's tool as an automatic organizing aid.

Read the Chronicle of Higher Education article about Westbrook and Smith and their audio detective work

Learn more about Ogihara, and read about his software in his University's news release