|contact us: firstname.lastname@example.org|
Music recognition with WIDI Recognition System step by step.
Let us take a particular example of the recognition of a musical mp3 record to a ready musical notation. Each step is illustrated by images and demos.
Each step of the tutorial has a detailed description, images and demos in the Flash format. You can open them in a separate window by clicking thumbnail images.
The detailed description of how to work with the WIDI interface is given in separate paragraphs and highlighted by a smaller font. If you know the interface of the program well, you can skip these paragraphs. The names of buttons and dialog boxes are written in the bold font, while the names of menu items are written in italics. Some of these names are links that refer to the detailed descriptions of the corresponding dialog boxes.
All files mentioned in the tutorial can be downloaded. Links for downloading files are highlighted in blue. It is recommended to download at least the source mp3 file and the final MIDI and wne files.
If some parts of the tutorial seem unclear or poorly worded, we will be grateful if you tell us your comments and suggestions in the corresponding section of our forum. This link is also available at the end of each page of the tutorial.
Source audio file.
As an example, we selected a fragment of Por Unos Dolares Mas from the album Music of the Andes by Inka Wayra provided courtesy of the authors.Download the source mp3 file, 448 KB.
First of all, you should listen to the track we are going to recognize. We hear a quite fast guitar accompaniment through the entire piece of music. Flute comes in a few seconds after the track starts and another wind instrument replaces it in the second part – it is Chilli, Andes inhabitants' traditional instrument. In this case, it is important to mention that both wind instruments are monophonic.
Recognition with the default parameters.
First, let us try to recognize the track without changing any parameters. If you have not changed the recognition parameters, it is enough just to start the recognition process, otherwise use the Default button in the Recognition Settings and Equalizer dialog boxes.
Open the Recognition Settings dialog box by selecting the Options->Recognition Settings menu item or by pressing the button on the toolbar in the main window. Click Default, than خت.
To open the Eqializer dialog box, use the Options->Adjust Eqializer menu item or the button on the toolbar. By default, all equalizer bars are set to maximum. If the current equalizer settings are different, use the Default button. Use the خت button to close Equalizer.
After you set the parameters, make sure that the entire track is selected (the start and end fields of selection are set to 0 and 57.13 seconds respectively) and click the button on the Wave window toolbar.
After the recognition process is over, the TrueTone editor window will appear with the recognition result. You can see this window in figure 1.
Note: Unlike in version 3.3, the MIDI Window appears in version 3.2 by default. To open the TrueTone editor, you should click the Edit MIDI data in TrueTone Editor button in it or just use the Ctrl+Enter key combination.
Adjusting the automatic recognition parameters.
First of all, look at the (Note Map) at the bottom of the window showing a reduced image of the entire track. You can clearly see smooth melody lines on the note map. It is a good sign, most probably the recognition result will need relatively little manual correction.
Flute comes in approximately at the fourth second and you can see its part as two quite smooth curves coming parallel at the interval of one octave. But a flute is a monophonic instrument while an octave is the interval between the main tone and the first (strongest) overtone. Therefore, the upper voice is an overtone and you should get rid of it. There are two ways to do it: remove the corresponding notes from the resulting file or change the recognition parameters so that a larger part of these notes does not appear there at all. The second method is more preferable because it allows you not to deal with parts of the spectrum that do not carry any musical information and increase the accuracy in the analysis of the remaining part. The highest note in the part of the flute is "Re" of third octave (D-5), and it occurs at the 22d second.
Besides, experience shows that it often makes sense to make the melody more prominent (usually it is the top voice) in a piece of music. You can easily do it all using the Equalizer.
Open the Eqializer and move the mouse pointer over the left equalizer bar. Press the left mouse button and move the pointer to the very bottom holding down the mouse button (the bar will be reduced to 0). Move it to the right after that until you reach the note "La" of first octave (A-1). The red dot and the text at the top of the Equalizer show the current note. Do the same with high frequencies: from the right bar to D#-5. After that draw a smooth curve on the rest of the spectrum making high frequencies more prominent.
The following Flash demo illustrates how to adjust the Equalizer:
Also, you can download a file with these Equalizer settings (with the .weqp extension) and load it into the Equalizer using the Import button.
Download a file with the Equalizer settings.
Comments about the text? Post them on the forum.