Products Download Order Support Site Map
contact us: info@widisoft.com
<< Previous page

After you repeat the recognition with the new Equalizer settings, you will get the result shown in figure 3 where you can see the first four seconds of the spectrogram (the image is reduced two times).

Figure 3

You can see from this spectrogram that there are quite a lot of unnecessary notes in it. To avoid it, you should understand where they are coming from. When a new note starts, there are quite a lot of atonal frequencies, i.e. frequencies that are not divisible by the main tone. It is neither the main tone, nor overtones. Since there is a resonator set for certain frequencies in any instrument (strings on a guitar, a cavity in a wind instrument and so on), all unnecessary frequencies fade quite fast. Therefore, these unnecessary frequency components sound for a quite short time. Besides, these frequencies sound much quieter than the main tone. So, most notes that are to be discarded from recognition are short and quiet.

Figure 4

One of the most important recognition parameters is the number of notes sounding simultaneously. The WIDI algorithm separates them by volume the number of the loudest notes specified in the settings is selected, the rest are discarded. Since the overtones of some notes may sometimes be louder than the main frequencies of other notes (as it happened with the overtone of flute in this piece), this parameter should be set to a bit greater value than the actual number of voices. After you listen to the source record once again, you will make sure that there are about 5 voices in it 4 guitar strings and the flute can be heard simultaneously. Set the polyphony value to 6 as it is shown in figure 4.

Select the Pitch->MIDI tab in the Recognition Settings dialog box and set the polyphony value to 6.

Figure 5

Repeat recognition and see the spectrogram. Now look at the part from 4 to 7.5 seconds. You can see that the number of unnecessary notes has decreased, but they are still rather numerous. That is why you should increase the minimal note duration. Take a look at this part of the spectrogram and select several wrong short notes on it. Set the minimal duration to a value that is a bit larger than the duration of these unnecessary notes. Figure 5 shows the recognition result with several unnecessary notes encircled with red and with their duration displayed. Set the minimal note duration to 65 ms (figure 6).

Figure 6

Select the Pitch->MIDI tab in the Recognition Settings dialog box and set the minimal note duration to 65 ms.
After you repeat the recognition, you can see that all encircled unnecessary notes have disappeared (figure 7).

Figure 7

Now pay attention to the synchronization of notes with different pitch (figure 8). The two selected notes located at about 44th second must sound simultaneously, but the bass note lags behind.

Figure 8

It is because the onset of bass notes is not defined precisely enough, besides, in most cases bass notes shift to the back or to the front depending on the recognition algorithm. The systematic shift of bass notes can be corrected by specifying the recognition parameters. You can find the detailed description of how to do it in the following insertion illustrated with figure 9. This operation improves the synchronization of notes with different pitches. The synchronization of notes with the audio file will be described later.

Figure 9

Open the Recognition Settings dialog box and select the Wave->Pitch tab. Move the (Time Shift of Bass) slider several points right. It is not always that you manage to specify the right value at once in this case the first shift was 4 points, but it was not enough. The shift was increased to 7 points and the difference between the onsets of notes decreased from 60 to 20 milliseconds.
Figure 10 shows the recognition result after the bass shift parameter has been changed.

Figure 10

The last step of adjusting the recognition parameters includes limiting the velocity (this features was introduced only in version 3.3). Using the method similar to the one that was used to determine the minimal note duration, select any unnecessary notes and see their velocity. In this case the minimal value is set to 25. Besides, if you pay attention to the way notes are divided by tracks (to do it, it is useful to select notes in the Track Editor using the A button on each track), you will see that tracks completely do not coincide with voices. That is why disable separating tracks by volume as well.

Figure 11

Open the Recognition Settings, switch to the Pitch->MIDI tab and set the Minimal Note Velocity parameter to 25. Besides, clear the Separate Tracks by Volume checkbox.

Automatic recognition result.

You can download the final result of all the above operations using the following link:
Download the final MIDI file.
Besides, you can download the configuration file that includes the recognition and Equalizer settings:
Download a file with the Recognition settings.



<< Previous     Next >>

Comments about the text? Post them on the forum.

Table of contents:
home | samples | presets mp3 to midi, wave to MIDI