A machine that knows a musical era within three notes
Editor’s note: This article is by Computational Biologist
Guillermo Cecchi of IBM Research.
of Western Music*
- 1700-1770 Baroque Era
- 1770-1830 Classical Era
- 1830-1900 Romantic Era
- 1900 - Post-Romantic Era
-- available on Peachnote
Pablo Rodriguez Zivic from the University of Buenos Aires, Favio
Shifres from the University of La Plata and I developed an algorithm that can
identify the Baroque, Classical and Romantic periods – after “hearing” only
added these near-three centuries and 20,000 songs worth of Western music periods
to its online database. Pablo tested our algorithm on the transcribed music
from Peachnote. Our machine then discovered patterns
across the songs down to the semi-tones or notes. The patterns learned could
now possibly extend to human speech or “humming.”
Visualization of the
clusters for the conditional distribution of melodic intervals. Each shaded
area corresponds to a different cluster, and its corresponding line represents
the proportion of years assigned to it within a 10-y smoothing window. Vertical
dashed lines correspond to the approximate boundaries between Baroque, Classical,
and Romantic periods. A fifth cluster was removed because it was a noise
cluster with only three elements.
Can you identify your favorite song within three consecutive
notes from any point within the song? In just three semi-tones, say “A,”
followed by “C,” followed by “F,” a pattern emerged that the machine accurately
matched to an entire era of music.
For example, we identifies late Baroque factors by the high
frequency of adjacent notes, such as “C” is often followed by “F” and “G,” and
then by “B” and “D.” The tune “Happy Birthday” follows this pattern:
[hap/C]-[py/C]-[birth/D]-[day/C]-[to/F]-[you/E]. As the tonal music developed
over time, the corresponding patterns of note combinations became more complex.
And while we did not focus on this aspect in the present
work, we have unpublished results identifying many of the composers, such as
Bach, Mozart and Beethoven.
Computers can do more
than identify music. Pablo Zivic has also developed a computationally creative
machine that can produce an era-based, but unique piece of music.
From music to speech
The structure of three consecutive semi-tones determines the
way we hear and perceive music. Simple factors, such as the time between the
three notes, the tone between the notes, and the order of the notes account for
identifiable trends within the eras of music.
We could perform this study because of the available data –
the “big data” of music. This same approach could identify other patterns in other
sounds, namely our speech. Computer algorithms can already identify speech
patterns in the early stages of Parkinson’s
disease through a recorded phone interview.
Doctors know that the vocal chords are affected early on in
the onset of Parkinson’s. We want to push the ability to identify the
combination of sounds – like the three notes in our Western music study – to
make an even earlier diagnosis.
With more-comprehensive speech data, we could uncover
patterns in other diseases. We're now using our tool to study past individual
cases to tease out features and patterns in the way people with a particular
psychiatric disorder speak or even “hum” music. From this, we could come up
with models to explain the behavior.
Perhaps in the future, other disorders can be identified
through simple, non-invasive verbal tests.
Labels: computational biology, machine learning, music, parkinson's