Music professor Michael Casey may have broken from the typical approach to a Thayer School of Engineering lecture when he asked audience members to make conversation while listening to the chorus of Madonna's "Lucky Star." Casey played the song to demonstrate advances in audio identification software in his lecture "From Vinyl to YouTube: Engineering the 21st Century Music Industry" in Spanos Auditorium.
Casey's lecture described the evolution of music technology since 1900, noting the development of phonographs, vinyl records and synthesizers, as well as the Synclavier, a digital synthesizer constructed by researchers at the Thayer School of Engineering in 1976.
As the availability of music has exploded over the last decade, so has technology used to record and identify music evolved, according to Casey, who chairs the music department and heads the Bregman Music and Audio Research Studio, a lab that explores connections between music and neuroscience.
In his lecture, Casey discussed the recent expansion and development of audio identification software. Using digital signal processing, smart phone applications such as Shazam are able to correctly identify songs even when they are played in the presence of other noise, a process Casey demonstrated by with his cell phone during the Madonna song.
These applications use conventional hashing techniques which convert several pieces of data into one to locate an item in a database, Casey said. The software charts recording characteristics such as pitch and drumline and creates a hash slot unique to that pair of characteristics. It then searches through the hash slots to find a match for the recording being played. If the slots match, the track is identified.
Music identification applications are effective even with outside interference, Casey said. In addition to working through ambient noise, applications can identify songs that have been distorted through a media player, even if the recording's time and frequency scaling have been altered.
"Applications are robust to additive interference, such as when you're in a bar or outdoors," Casey said.
The software can analyze audio so specifically that only an exact match sequence can be identified as the original, according to Casey. The software will not confuse songs with similar audio characteristics, even if they are two versions of the same song, he said.
Despite the success of music identification software, transcribing musical notes from a recording is far more difficult, Casey said.
"When you hear a drum break, it sounds like only one instrument is playing, but it may actually be one bass, one cymbal, one high hat the timbre might be changing," he said. "Rhythm transcription is not exact if it's written in multiple lines, it will be difficult to extract reliably from an engineering perspective."
Instead of attempting to transcribe multiple layers of melody, music information retrievers use spectrograms to graph signals that humans hear. The programs graph frequencies or energy waves to represent different pitch classes and the chords being played. With a greater sample length, these graphs can show how the music changes over time and extract harmonies, according to Casey.
The scale of media itself has also presented a problem for music engineers, Casey said. Music is available in a variety of forms, from personal music devices holding 10,000 songs to music-sharing services that offer tens of millions.
"The take home message is that there's a very large amount of data and the music industry has to tackle this on an immense scale," he said.
Because of the sheer number of songs in existence, music identification applications must own immense amounts of music, as they can only identify songs that exist within their own databases, Casey said.
"[Casey's talk] was kind of out of the ordinary compared to what we're used to getting here," said Evert Holwerda, a graduate student at Thayer, adding that lectures at Thayer are usually more technical in nature. "But it turned out to be a very technological, science-driven presentation, even though [Casey] came from an arts background."
