To convert a wave file to a midi file just open the wave file,
analyse it and save the midi file! Once a midi file has been saved
it may also be played.
If it was really that easy to get good results then all those
parameters would not be needed. To get the best results some fiddling
will be required.
|
Tuning |
The expected frequencies are based on A5=440Hz. This can be altered
by up to 50 cents each way (half a tone) to produce a better
fit with any possibly out-of-tune recordings. After a piece has
been partly or fully analysed the automatic facility may be used
to detect the most likely tuning. This is the only parameter
which requires a full re-analysis after it has been altered.
All other parameters only require the piece to be re-filtered. |
|
Polyphony |
Not quite what is normally meant by polyphony. This is the number
of new notes that can start at once. Current notes can carry
on playing. |
|
Minimum chord component |
When several notes are playing simultaneously, any notes with
an amplitude lower than a percentage of the loudest current note
will be filtered out. |
|
Neighbouring notes threshold |
When two adjacent notes are played then the amplitude of the
weaker must be at least a certain percentage of that of the stronger,
otherwise it is filtered out. This is mainly to help with low
note recognition, where playing just one note produces a frequency
spread of several notes during analysis. |
|
Overall minimum component |
The amplitude of the weakest note saved in the midi file as a
percentage of the loudest note in the whole piece. |
|
Amplitude fluctuation |
The frequency analysis does not lead to a smoothly decaying amplitude
of each note because of the influence of other notes. If the
amplitude at each analysis segment falls within a band around
a running average then it is smoothed automatically. This helps
with further analysis. For a new note to start while the same
note frequency is still playing (ie repeating a note while the
sustain pedal is in use on a piano) the new note amplitude must
be 10% above the allowed fluctuation of the running average over
the last few samples. |
|
Phase analysis |
Another low note recognition routine. When two adjacent notes
are detected then an additional analysis may be used based on
measuring the phase of each note throughout the sample. This
is a slow procedure, but because it is only used for pairs of
notes (after the neighbouring notes threshold filter has been
used) it only adds about 20-30% to the time taken for analysis.
However, during any very quiet parts of the music such as any
silence at the start and end, it will be used for almost each
note and its neighbour, greatly slowing down the analysis during
the silence. In practice it is of little value and will probably
be removed before the final version is released. |
|
Single sample noise removal |
Strong harmonics may occasionally give rise to spurious notes.
Checking this option will remove any notes which are only present
for a single sample interval. |
|
Notes start loudest |
Certain instruments, such as any keyboard instrument, have control
over the volume of a note at the start, after which the volume
can only decay. Enabling this option will enforce this, compensating
for any possible weak starts from the analysis. |
|
Draw Spectrum Analyser |
Pretty, but unnecessary. During the filtering stage after a full
analysis the frequency spectrum can be displayed throughout the
piece as the note map is built. During re-filtering without a
full analysis the spectrum is never displayed because of the
time delay involved. |