Specmurt Anasylis | ||||||||||||||||||||||||||||||||||||||||||||||||
"Specmurt Anasylis" is an unique method invented by
Prof. S. Sagayama in 2003 that enables "pitch likelihood
distribution visualization" unlike the standard "pitch
estimation" from a single channel multi-pitch audio signal. It is
a frame-wise signal processing method like short-time spectrum
analysis. Opposed to "Cepstrum Alanysis" which is the (inverse)
Fourier transform of short-time power spectrum with log-scaled
magnitude, proposed by Tukey et al. in 1964, we defined "Specmurt
Anasylis" as the (inverse) Fourier transform of short-time power
spectrum with log-scaled frequency. Division of the power spectrum
by an assumed "common harmonic structure" in the specmurt domain
gives the fundamental frequency distribution. In the following
sample data, we used an iterative algorithm, proposed by H. Kameoka
and S. Saito in 2004, that chooses proper characteristics of the
"common harmonic structure" [Kameoka2004MUS08] followed by an HMM
to trace multiple pitch contours for converting from the output
distribution to MIDI
data, though this improvement is not mentioned in the SAPA2004
paper.
Keywords: multipitch, overtones, harmonics, automatic transcription, multitone analysis, fundamental frequencies, spectrum analysis
| ||||||||||||||||||||||||||||||||||||||||||||||||
Bibliography | ||||||||||||||||||||||||||||||||||||||||||||||||
This idea and preliminary results were first
published in Japanese [Takahashi2003MUS12] and in English
[Sagayama2004SAPA10]. Iterative estimation of the optimal common
harmonic structure was first included in [Kameoka2004MUS08] written
in Japanese. Its English version is under preparation.
| ||||||||||||||||||||||||||||||||||||||||||||||||
A Sample of pitch visualization of audio data | ||||||||||||||||||||||||||||||||||||||||||||||||
The 5-th voice (flute) entering J. S. Bach's 6-voice Ricercare
from "Musical Offering" (BWV1079) excerpted from the RWC Music
Database.
| ||||||||||||||||||||||||||||||||||||||||||||||||
Preliminary results of MIDI conversion from audio data | ||||||||||||||||||||||||||||||||||||||||||||||||
|
[ Back to Lab Home ]