Information Technology Reference
In-Depth Information
In this contest it is difficult to point out the
accuracy rate of the various algorithms because
some widespread dataset and common method-
ological evaluation routine are not yet accepted.
This situation is partly due to the choice of dataset,
which often depends on the goals of the system
(Dixon, 2001).
Various models have been proposed in or-
der to extract the beat from performance data.
The primary distinction we want to point out
is between real-time and batch algorithms. For
example automatic accompaniment systems have
to use real-time algorithms. Transcription and
analysis software tends to process data off-line,
because rhythmically ambiguous sections can
be frequently determined analyzing all the beat
information found in the song. Thus the choice
between real-time and off-line systems it is directly
related to algorithm aim.
Actually the beat tracking system works on a
two-stage model. The first stage is an onset de-
tector, the second one is an interpretative system,
which gets the onset detector output and tries to
understand the tempo of the song and the correct
beat position.
For music with drums to develop an onset
detector system the simplest way is to high pass
the signal and then clustering the filtered signal
and introduce a threshold to the cluster energies.
Obviously this trivial algorithm does not work
very well, but it permits understanding that a
drums beat has a spectral frequency with rich
high components.
With the interpretative system, it is possible
to find many different solutions like the agents
model (Dixon, 2001; Goto, 2001), or probabilistic
systems (Sethares, Sethares, & Morris, 2005)
and so on.
We will present two different solutions to
the interpretative problem that use the agents
model.
In the next sections, four different tempo-
beat tracking algorithms (one solely dedicated to
tempo tracking) will be described. The algorithms
(Dixon 2001; Goto, 2001) work in PCM format,
the algorithms presented in (D'Aguanno, Haus,
& Vercellesi, 2006;Wang, & Vilernmo, 2001) are
dedicated to MP3 standard.
The Goto Algorithm
The first algorithm was developed by Masataka
and Goto (2001). This algorithm is based on the
previous Goto and Marauroka (1998; 1999) works,
one for music with drums and the other for mu-
sic without drums. In Figure 23 is presented the
algorithm scheme.
This algorithm describes a real-time beat-
tracking system that can deal with the audio
signals of popular-music compact discs in real
time regardless of whether or not those signals
contain drum sounds. The system can recognize
the hierarchical beat structure comprising the
quarter-note level (almost regularly spaced beat
times), the half-note level, and the measure level
(Goto, 2001)
The algorithm consists of two components: the
first one extracts the musical elements from audio
signals; the second component tries to understand
the beat structure.
The first step detects three kinds of musical
elements as the beat tracking cues:
•
Onset times
•
Chord changes
•
Drum patterns
These elements are extracted from the fre-
quency spectrum calculated with the FFT (1024
samples) of the input (16 bit / 22.05 kHz) using
the Hanning window. The frequency spectrum is
subdivided into seven critical bands. The onset
times can be detected by a frequency analysis
process that takes into account the rapidity of
an increase in power and the power present in
nearby time frequency bands. The results of this
algorithm are stored in an onset-time vector. By
using autocorrelation and crosscorrelation of the
Search WWH ::
Custom Search