Computers' improve their sense of rhythm
Some of Antti Eronen's methods are so efficient that they can
process millions of musical pieces and are thus paving the way
for new commercial Web and mobile phone applications.
Senior Researcher Antti Eronen plays the song Ain't no sunshine by Kenny Rogers on his laptop. The computer taps to the rhythm - in perfect unison. Santana's Black Magic Woman leaves the computer puzzled at first but it soon catches up with the beat. The computer is also quick to recognize the chorus of the song.
The laptop owes its technical prowess to methods developed by Eronen. These methods can be applied to rhythm analysis, the identification of musical instruments and the chorus, and the classification of acoustic environments based on digital audio signals. Eronen has conducted his research within the framework of joint projects between Nokia Research Center and the Department of Signal Processing at TUT.
Computers know rock and pop
Computers are able to produce exact results when analyzing the tempo and chorus of pop and rock music. The applications could be used, for example, to find certain type of music from the user's own MP3 collection or online music catalogues.
"Customers looking for slow or fast paced music from an online music store could start listening to the song straight from the chorus, which is usually the most representative part of any piece of music," says Eronen.
With the advent and consequent popularity of MP3 players, research in the field of musical signals has increased rapidly. Although still a relatively young field, some of the methods are already reliable enough for commercial use. Examples of potential application areas include music identification, music retrieval by humming, validation of music recommendations, categorization by genre, mood or artist, and automatic transcription.
Eronen considers tempo analysis the most mature application area for commercial use but specifies that it is not a new research topic:
"BPM estimators have been a standard feature in the musical equipment for professionals and DJs for a long time now, but many of the methods work well only with music that has a strong drumbeat. The new methods, however, are also capable of producing quite a reliable analysis of the tempo of music without drums."
Computer no match for the human ear
A computer's sense of rhythm is based on metric analysis. In music, a metre refers to a hierarchical structure of beats and pulses. For example, in 4/4 time each measure is divided into four beats that are further divided into shorter pulses. The perception of a beat emerges as a result of recurring events in the song, especially of acoustic attacks such as notes, drumbeats or rests.
The user downloads music samples, which have different tempo and the basic beat notated by hand, onto the computer. From these samples the computer measures the attacks, examines their periodicity and attempts to model what the different metres are like. Then the machine tries to identify the tempo and bar-lines of other songs by comparing them with the models.
According to Eronen, computers are quite good at analyzing the average tempo of different types of music but bar-lines are a different matter. An accurate analysis of the structure and metre of music is also difficult, not to mention the automatic classification of musical instruments.
The human ear can identify different types of instruments based on, for example, pitch, sound duration, volume, and tone quality. In Eronen's view the challenge is to get a straightforward computer to identify tones from polyphonic music. Another obstacle is that each individual instrument has its own unique timbre, which is also influenced by the player's technique and the acoustic environment.
"The main problem a computer faces is related to generalization: it should be able to identify different sounding violins and guitars as violins and guitars irrespective of the player and the environment. Further research will likely shed more light on this dilemma, too."