I'm the creator of Eldoraudio, a suite of tools focused on audio-to-MIDI conversion. I've spent the last couple of years going deep on automatic music transcription, starting with piano-to-MIDI using a custom-trained model, and currently building out guitar transcription.
One thing I keep running into is how differently the MIDI community uses transcription output. Some people want note-perfect MIDI for sheet music generation, others want something that captures the performance feel even at the cost of some accuracy, and others are piping it straight into a DAW where timing quantization matters more than dynamics.
Curious what this community's experience has been: what's the most frustrating gap you've hit between what audio-to-MIDI tools produce today and what you actually need the output to do? And which instrument has felt most underserved?
The biggest gap is getting clean, editable MIDI without constant manual cleanup—especially around timing, overlapping notes, and articulations. Guitar still feels the most underserved, particularly for chords, bends, slides, and expressive playing.
I have a rather fringe case of trying to replicate video game sound effects in MIDI. Some of these are synthesized speech. I have thus far not been able to find any utility that gives me a satisfactory recreation of said sound effects/voices. I realize voice will be extremely difficult, but hoping one day to to find something that comes close enough where I can choose an instrument that might make it sound "comprehensible". One example is the beginning music from the SNES game Star Fox, where a key part of the music track is a robotic voice saying "Emergency, Emergency" along with some other brief dialog. On a previous soundtrack I did, I had to settle for a mediocre recreation of some other robotic sound effects that I just couldn't get any utility to recreate well. (The original music is sample based, and the samples just didn't MIDI-ise well enough for my tastes)
For me, I do not care how the final MIDI "looks", as I have to take it and finesse it in to the MIDI files I am working on. Options to specify things like TPQN, tempo, etc would be beneficial, though, as I could then match the audio-to-MIDI conversion to my existing file. And options to allow, disallow, or prefer certain commands to generate the final result (for example, use expression instead of volume to control dynamics) would also be nice to have.
Hi Jason, great feedback on the TPQN and tempo matching. That was actually one of the most requested features, and I'm happy to say BPM specification is already supported on Eldoraudio's piano audio to MIDI converter.