All TS-AudioToMIDI recognition algorithms except Mono Correlator are based on a set of oscillatory circuits we call "sensors". There are 87 sensors in the set, by number of keys in a standard piano keyboard. Each sensor is tuned to one of the standard note frequencies. Corellation of sensors is controlled by Selectivity parameter. By changing Selectivity you change resonance curve shape (it is drawn in yellow color on Selectivity window). The set of sensors gives spectrogram as output.
Next stage is spectrogram analysis. First, spectrum is amplified by Equalizer values. The result is processed by one of the note detection algorithms. All algorithms except Mono Correlator use Harmonic Model as one of the most important components.
Each musical note nave many spectral components. The lowest one is fundamental tone, it's frequency is perceived as note frequency, and other components, called overtones, have frequencies at 12, 19, 24 and so on semi-tomes higher than base tone. Overtones determine the timble of note. Each instrument has it's own harmonic components amplitude ratio. TS-AudioToMIDI assumes that all notes to be detected have inter-component ratio as it is set in Harmonic model window.
Except harmonics ratio for each note two others features are analyzed. They are: amplitude (if amplitude is less than Threshold value, note is rejected) and pitch shape, controlled by Sharpness parameter.
After this each note is compared with Minimal Note duration value, and all notes forbidden in Filter window are filtered out.
In contrast, Mono Correlator recognition algorithm is based on correlation function. Correlation analysis allows to trace a note during all the time it sounds. Mono Correlator does not use Harmonic model, however Minimal Note and Pause duration values as well as Filter settings are taken into account.
Please note, that real-time recognition does not take into account Minimal Note and Minimal Pause duration values, that leads to the difference between what you hear during real-time recognition and what is recordered by built-in MIDI sequencer. To avoid this difference set Minimal Note and Minimal Pause durations to zero.
Related topics:
Choosing Recognition Algorithm
Tuning up Selectivity Window
|
