Separated references before MIDI guitar transcription?

General
1

I have a MIDI guitar and MIDI bass practice and recording question.

When checking whether vocal/instrumental separation helps before attempting MIDI transcription or tracking work, do you find a private vocal/instrumental separation useful as a first listening aid, or does it create more problems than it solves?

I am not talking about posting extracted stems, sharing copyrighted tracks, linking a tool, or using a separated file as a release asset. I mean a private listening step where the player has permission to work with the audio and wants to decide whether the split helps.

The cautious workflow I would use is:

  1. Use only audio I made or have permission to work with.
  2. Keep the original recording as the reference.
  3. Preview the separated vocal and instrumental result before trusting it.
  4. Listen for artifacts in the busiest section of the song or demo.
  5. Treat the separated result as scratch material, not proof.
  6. Avoid uploading files, screenshots, or temporary processing links into the thread.

For MIDI guitar, MIDI bass, pitch tracking, transcription prep, practice recordings, and arrangement analysis, where is the line between a useful practice aid and a misleading reference? separation artifacts can confuse pitch tracking or make timing feel cleaner than it is.

Separation quality varies by track, mix, source audio quality, and model/provider behavior; users should only upload audio they have rights to process.

2

an issue you face is that extracted stems will likely include effects like reverb which will adversely affect note detection.

i’ve run separated stems thru mg3 and although the accuracy was close to ‘useful practice aid’ it did mislead at times.

the track involved was an older reggae song, a relatively clean recording.

timing-wise, the stem extractor i used gave me files which varied slightly in total length and did not match up perfectly.

accuracy in pitch and timing when transcribing from stems will probably range between 90% and 97%.

mg3 does not do pitch detection on vocal tracks.

3

Yeah, that makes sense to me.

I’d be a bit wary of the timing side of things too. Even if the separated part sounds clearer, if the file is slightly stretched, smeared, or not lining up perfectly with the original, it could give you the wrong idea about where the notes actually sit.

For practice or working things out by ear, I can definitely see it being useful. Pull the part forward, get a rough idea of what’s going on, then go back to the original recording and check it properly there.

For MG3 or pitch-tracking, I’d be more careful. Reverb, weird separation artifacts, softened attacks, missing note starts, or timing that feels a bit too tidy could all make the result look more reliable than it really is.

So I’d probably treat it like EQ, slowdown, looping, or any other transcription trick: handy as a magnifying glass, but I wouldn’t make it the source of truth.