Open timtensor opened 8 months ago
Hi! Can you please share your code and the midi file you used in your test ? We'll have a look :)
Hi thanks for the reply . I actually used a standard salsa piano snippet ( montuno ) beat which I don't have the midi at the moment .
I was wondering this implementation is more like conditioning and interpolation is it ? I think is it similar to Magenta ?
Hi, without the midi file I can't be a great help, we recently fix some bugs on midi file ingestion so maybe it did the trick (1.1.8). For your question, at the end it's really an autoregressive architecture (LLAMA2 architecture based), the innovation lies in that the note pitches are written relative to the current chord. That's why chords are controllable. Maybe this article can help : https://musiclang.github.io/tokenizer/
Hi i dont think i can add the midi file here . Is there any way to upload the midi file ? Is there, any also some youtube turotials that one can follow as well ?
Hi @floriangardin - just a curious follow up question . which code snippet should i use to experiment with my own midi file ? Kind of like song continuation . Suppose I have a midi file name 'test1.midi'
i used a reference midi file to generate variations. It does not seem to have worked. Does it only work for select midi songs. I used a simple piano midi file as a reference file . Perhaps i am wrong somewhere here