Multi-Instrument MIDI Transcription (MIROS) · TelkNet

MIDI / transcriptionGPU workload

Multi-Instrument MIDI Transcription (MIROS)

Users who need editable MIDI from a full mix quickly

MIROS-generated MIDI for multi-instrument transcription review and DAW cleanup

3 credits/useAudio file inputMax 50 MBEstimated 180-900 seconds

Input

Full songs, ensembles, or multi-instrument demos through the official MIROS AMT path

Audio formats

mp3wavflacoggm4aaac

Output

MIROS-generated MIDI for multi-instrument transcription review and DAW cleanup

Best for

Users who need editable MIDI from a full mix quickly

Models, parameters, and sources

Multi-Instrument MIDI Transcription (MIROS) runs the official MIROS MusicFM multi-decoder route for full-mix multi-instrument AMT.

MIROS

MIROS is another multi-instrument AMT backend and can be used as an alternative to YourMT3.

amt-os/ai4m-miros Official leaderboard 2025 AMT Challenge paper

Public evidence: MIROS ranked first in the 2025 AMT Challenge table cited here.
Scope: multi-instrument AMT for classical/polyphonic material, not audio stem separation.
Runtime: exposed as an alternative backend where MIROS source and weights are available.

Model comparison

Model

Role

Why it is used

Output impact

ModelMIROS

amt-os/ai4m-miros 2025 AMT Challenge paper Official leaderboard

RoleChallenge-ranked multi-instrument AMT

Why it is usedUse when the MIROS runtime is available and maximum multi-instrument recognition quality is preferred.

Output impactMulti-track MIDI

ModelYourMT3+ YPTF.MoE+Multi

mimbres/YourMT3 YourMT3 arXiv 2025 AMT Challenge paper Official leaderboard

RoleDefault robust multi-instrument AMT

Why it is usedStable TelkNet path with exposed checkpoint, quantization, and post-processing controls.

Output impactMulti-track MIDI draft

ModelMT3 baseline

MT3 arXiv 2025 AMT Challenge paper Official leaderboard

RoleHistorical baseline

Why it is usedKept as public benchmark context for the MT3 family, not as the recommended TelkNet path.

Output impactBaseline comparison only

Public benchmark chart

Scores are public benchmark or source-reference rows. They are not a fresh TelkNet run on the uploaded file.

2025 AMT Challenge ranking

Official leaderboard

Multi-instrument transcription challenge rows. Higher F1 is better; shown as public challenge evidence, not a TelkNet re-run.

MIROS#1

F1 (%) 59.98Precision (%) 65.58Recall (%) 57.24Overlap (%) 73.91Runtime (source unit) 22.05

MIROS is the top public row in the cited 2025 AMT Challenge table.

YourMT3-YPTF-MoE-M#2

F1 (%) 59.38Precision (%) 60.10Recall (%) 58.88Overlap (%) 73.05Runtime (source unit) 12.60

MT3 baseline#5

F1 (%) 39.32

MT3 is included as the historical baseline named in the challenge paper.

Paper and benchmark notes

Advancing Multi-Instrument Music Transcription: Results from the 2025 AMT Challenge

arXiv / OpenReview - 2026

The 2025 AMT Challenge paper reports current multi-instrument challenge rankings including MIROS, YourMT3, MT3, and Basic Pitch rows.

arXiv:2603.27528

Official / repositories

amt-os/ai4m-miros Official leaderboard

Papers / technical notes

2025 AMT Challenge paper

Parameter guide

Fixed adapter: the tool_id selects the dedicated model; generic YourMT3 parameters are not exposed.

Use cases

Transcribe full demos and analyze harmony
Turn idea recordings into editable MIDI
Start with multi-track drafts, then refine by instrument in a DAW

Workflow

1Upload a full audio file with a stable tempo
2Wait for multi-instrument detection to produce MIDI
3Import the result into your DAW to fix velocity, quantization, and tracks

Pre-flight checklist

Keep audio with a stable tempo and avoid long silent stretches
More complex arrangements need more manual note correction afterwards
For solo piano, switch to a piano-specialist model instead

Login required to use this tool

Browsing tool info is always free - no account needed.