Buy It Now
About MP3 to MIDI
This review on
intelliScore Wave to MIDI Converter
(translated from Dutch) was originally published at http://www.dosgg.nl/.
From Wave to MIDI
WAVE files contain raster-sound, MIDI
contain vector sound: commands to create (raster-)sound.
If you let this fact act in upon you, you
can yourself realize how at least to go from MIDI to WAVE: feed the
commands from the MIDI file to a synthesizer, and record the sound that
resounds with the settings chosen. For example, see the documentation with
a Sound Blaster.
You then also see that it will be
difficult to deduce from the raster-sound, what commands are needed to
create it. (Have you ever tried to vectorize a photograph with e.g. Corel
Trace?) That problem does not date from the computer age. There is enough
sheet music in circulation that has been obtained through transcription:
somebody has carefully written down in musical notation what he heard
played. The computer aggravates the question to the extent, that the
"note-taker" must indicate how this recognition occurs - and
that presupposes becoming aware of a "well, just!" process. On
the other hand, you need to write a program to recognize notes only once.
Transcription is not unequivocal: even when listening to a perfectly
transparent recording, you can doubt about when a tone begins, and
especially when it stops. That depends on the characteristics of the
instrument (bowing is much less definite than fingering) and on the
acoustics (reverb). This implies for MIDI: when must the Note On command
be sent, when Note Off? And then there are keyboard-instruments with a
sustain pedal… But moreover there are questions like: are these eighth
notes with eighth rests in between, or are they crotchets played staccato?
It doesn't matter for MIDI-files that will only be played, but it does for
MIDI files that will be used for printing sheet music.
There is definitely a need for
transcription programs, e.g. to be able to review "what I actually
did". Or as a drastic way to denoise historic recordings. Or to add
the play of non-MIDI instruments to a MIDI file yet. Therefore, such
software is being programmed everywhere. I got the impression that
predominantly Russians (rich on mathematics, poor on money) had thrown
themselves on the issue. Yet I have an American program now. As for now,
one should judge these programs on what they do correctly already, not to
what mistakes they still make. Or yourself prove that it can be better,
e.g. by making drums scores recognized as well, for these are now filtered
From intelliScore there exist a monophonic
and a polyphonic version. According to me, the monophonic one is more of a
marketing product. It is worked upon hard: from version 1 to version 3
within a year. The program sets itself apart (from the two Russian ones
that I have tried, that is) by the extent of involving the user in the
process and by the strive to serve both sequencers and notation programs.
For intelliScore you need a reasonable
system (a CPU with MMC-instructions, and Windows 95 or higher). Despite
this also patience, for the recognition of a piece lasts the normal
playing time at least, many times that if you are unfortunate. And
tenacity, for you must be prepared to repeat that recognition several
times. Musical insight too, to adjust the settings meaningfully and to
complete the job in a sequencer or notation program afterwards.
IntelliScore acts like a funnel. You start
a project using ample assumptions, e.g. as to the tonal range. After each
recognition you can "tighten the bolts" for the next round.
(Herein lies the craftsmanship. Have you ever tightened the spokes of a
bicycle wheel?) If you consider yourself advanced, you have access to more
settings than as a novice. If the raw material has one sound (a piano, a
set of recorders), then you can choose a filter that takes the
characteristics of such an instrument into account. This shortens the
recognition time and improves the accuracy. As to me the very improvement
over version 2.
At last you have a product you consider
"I'll do the rest myself". (Tip: use serial numbers to be able
to compare the MIDI files of the recognition rounds.) That product is a
Type 0 MIDI file, or: all notes will be played back on the same channel
using the same sound. If you want to separate the voices, you'll have to
divide the commands over different tracks and MIDI channels. (Try to
distinguish the first violins from the second in an old wax roll recording…)
Finally save your work as a Type 1 MIDI file or in the custom format of
Finally a word about the method to analyze
sound (book-learning, I don't know anything myself). The standard approach
was described in 1822 by Frenchman Joseph Fourier: find frequencies that
are characteristic of certain events. Compare it to the reception by
FM-radio. Finding sine-shaped waves is a crude approach, however, if the
very waves are not sine-shaped. In the 1980s the wavelet-approach was
developed by Jean Morlet (France) and later Ingrid Daubechies (Belgium).
Here one starts from wave patterns rather than from frequencies and
overtones. In 1995, Ronald Coifman (USA) has polished a classical
recording from 1889(!) this way. According to the FAQ on
www.intelliscore.net, intelliScore uses a new way (or a smart mix), for
which patent has been applied for.
Chris Laarman, May 10,
(translation: May 14, 2000)
to reviews page