I have a dataset of midi songs and I would like to clean them as much as possible without losing much information. I really dont know much about music and I just started to work with midifiles so my knoledge is limited. What I have done so far is filtering all the events that seem irrelevant to me. The events I am keeping are:
- Note events
- Tempo events
- Control change events (only with control 64, which I think represents the pedal of the piano)
Another thing Ive done is to erase all the tempo events and set just one tempo event to 120bpms at the beggining by replacing the ticks of the other events according to the previous tempo events. Now I would like to be able to clean the control change events aswell.
Is there a way to modify the note events (maybe the duration) to achieve a similar effect to the control change event so that I can erase all the control change events (since most of their information would be present in the notes)?
What I want to achieve is a format that only encodes the note characteristics (time of appearance, duration, pitch, velocity) but be able to reproduce almost the same song.