Fermata Transcriber

Audio in, beat-aligned multi-instrument MIDI out, on a CPU. Every note is placed on the song's bars and beats with a General MIDI instrument per track and drums, so the result opens as a usable score in a DAW or notation program.

Try it, no sign-up

First 60 seconds of your upload · 3 songs a day · uploads are deleted after transcription

Piano roll of a transcribed song: several instruments and drums over 60 seconds — First 60 seconds of a Creative Commons recording, transcribed on 2 CPU threads in 17 seconds.

What you get

synced MIDI: notes on the beat grid, tempo follows the recording, so it lines up with the audio in a DAW
score MIDI: notes on the beat grid at one tempo with the meter, for notation programs
about 78M parameters: audio encoder (28M) and a hybrid Gated DeltaNet / attention decoder (50M) running on llama.cpp
a 150-second song takes about 135 seconds on a 4-core Intel N150 mini PC (8-bit weights)

Benchmarks

Note-level F1 (%), mir_eval, 50 ms onset tolerance, mean over pieces. MuScriptor was run by us on the same audio and scored with the same code. Higher is better; shaded = best in column. Fermata rows are v9, the version behind the trial.

MulTTiPop test · real pop, 385 songs	Onset	Offset	Exact (instrument)	Multi-inst.	Drums
MuScriptor small	35.8	17.8	15.2	17.3	30.3
MuScriptor medium	41.3	20.5	16.6	18.9	32.1
Fermata Transcriber	40.5	18.5	18.9	18.2	31.8

PHENICX-Anechoic · orchestra, 4 pieces	Onset	Offset	Multi-inst.
MuScriptor small	50.1	34.9	6.2
MuScriptor medium	54.4	39.7	0.0
Fermata Transcriber	61.0	31.2	22.8

Slakh2100 test · synthesized, 50 songs	Onset	Offset	Exact	Multi-inst.	Drums
MuScriptor small	50.9	23.8	17.7	21.1	39.7
MuScriptor medium	54.0	26.0	18.9	22.3	41.9
Fermata Transcriber	62.5	31.6	34.8	32.6	52.2

Slakh bass references were shifted down one octave to match the audio. Inference settings were chosen on these test sets (no held-out split), so treat gaps of about 1 point as noise.

Limitations

Instrument labels are the weakest part; pitch and rhythm are much more reliable than the GM program choice.
Note ends trail note starts. Vocals are not written as a melody track.
Bar lines follow the beat tracker, so rubato or meter changes can shift downbeats.