What tasks can Singing Voice to MIDI · Four Models (VocalParse 1.7B) help with?

MIDI / transcription; Compare VocalParse, GAME, ROSVOT and SOME for singing-note transcription, including score-aware, accompaniment-robust and fast paths.; A complete supported audio or video file; choose the model profile that matches the source and the result you need.; Turn a complete recording into editable notes, drum hits, tablature or a chord timeline.; Compare the model output with the original audio to review onsets, boundaries and source assignment.; Move MIDI and time-aligned tables into a DAW, notation editor or downstream analysis workflow.

MIDI / transcriptionGPU workload

Singing Voice to MIDI · Four Models

Q: Who is Singing Voice to MIDI · Four Models (VocalParse 1.7B) for?

MIDI / transcription Compare VocalParse, GAME, ROSVOT and SOME for singing-note transcription, including score-aware, accompaniment-robust and fast paths.

Q: Which model and keyword entities are connected with Singing Voice to MIDI · Four Models (VocalParse 1.7B)?

VocalParse 1.7B, GAME, ROSVOT, SOME, VocalParse source (pinned), VocalParse weights (pinned), SingCrawl dataset source, VocalParse paper

Compare VocalParse, GAME, ROSVOT and SOME for singing-note transcription, including score-aware, accompaniment-robust and fast paths.

5-20 credits/useFile inputMax 100 MBEstimated 45-1800 seconds

Input

A complete supported audio or video file; choose the model profile that matches the source and the result you need.

Supported formats

mp3wavflacoggm4aaacaifaiffmp4movmkv

Output

Compare VocalParse, GAME, ROSVOT and SOME for singing-note transcription, including score-aware, accompaniment-robust and fast paths.

Use case

Compare VocalParse, GAME, ROSVOT and SOME for singing-note transcription, including score-aware, accompaniment-robust and fast paths.

Models, parameters, and sources

Compare VocalParse, GAME, ROSVOT and SOME for singing-note transcription, including score-aware, accompaniment-robust and fast paths.

VocalParse 1.7BGAMEROSVOTSOMEVocalParse source (pinned)VocalParse weights (pinned)SingCrawl dataset sourceVocalParse paperGAME source (pinned)GAME v1.0.3 weightsGAME technical guide (pinned)GAME ONNX guide (pinned)

VocalParse 1.7B

Default path for structured, score-like vocal MIDI from cleaner singing; the fixed checkpoint has no official user-tunable inference parameters.

VocalParse source (pinned)VocalParse weights (pinned)SingCrawl dataset source VocalParse paper

Dedicated route: this tool uses the model path selected for this workflow.

GAME

GAME converts singing to MIDI with its official boundary, D3PM and estimation controls. Language defaults to auto/unspecified; Chinese is not forced.

GAME source (pinned)GAME v1.0.3 weights GAME technical guide (pinned)GAME ONNX guide (pinned)

Auto uses GAME's official universal/unspecified language ID 0. Select a specific language only when the sung language is known.
GAME boundary-segmentation confidence threshold.
GAME boundary search radius in seconds.
Starting noise time used by GAME’s official D3PM schedule.
Number of GAME D3PM denoising steps.
GAME note-estimation confidence threshold.
After inference, Beat This final0 writes a valid fixed tempo or beat-by-beat tempo map into the MIDI while preserving every event's position in seconds. If valid beat evidence cannot be produced, the task fails instead of delivering MIDI with the model's default tempo.

ROSVOT

Robust path for singing with noise or accompaniment leakage; the current official checkpoint is trained mainly on Mandarin singing.

ROSVOT source (pinned)ROSVOT project page ROSVOT paper

ROSVOT note-boundary confidence threshold.
ROSVOT word-boundary confidence threshold.
Enable ROSVOT’s official slur consistency check.

SOME

SOME is the fast, lightweight singing-to-MIDI route. It exposes no public model controls; optional MIDI tempo enhancement runs after inference.

SOME source (pinned)SOME baseline weights SOME technical guide (pinned)

After inference, Beat This final0 writes a valid fixed tempo or beat-by-beat tempo map into the MIDI while preserving every event's position in seconds. If valid beat evidence cannot be produced, the task fails instead of delivering MIDI with the model's default tempo.

Paper and benchmark notes

VocalParse: Towards Unified and Scalable Singing Voice Transcription with Large Audio Language Models

arXiv - 2026

Default path for structured, score-like vocal MIDI from cleaner singing; the fixed checkpoint has no official user-tunable inference parameters.

Paper

Robust Singing Voice Transcription Serves Synthesis

ACL - 2024

Robust path for singing with noise or accompaniment leakage; the current official checkpoint is trained mainly on Mandarin singing.

Paper

GAME: Generative Adaptive MIDI Extractor

Official technical guide - 2026

GAME converts singing to MIDI with its official boundary, D3PM and estimation controls. Language defaults to auto/unspecified; Chinese is not forced.

Technical guide (pinned)ONNX guide (pinned)v1.0.3 weights

SOME: Singing-Oriented MIDI Extractor

Official technical guide - 2023

SOME is the fast, lightweight singing-to-MIDI route. It exposes no public model controls; optional MIDI tempo enhancement runs after inference.

Technical guide (pinned)Baseline weights

Official / repositories

VocalParse source (pinned)VocalParse weights (pinned)SingCrawl dataset source GAME source (pinned)GAME v1.0.3 weights ROSVOT source (pinned)ROSVOT project page SOME source (pinned)SOME baseline weights

Papers / technical notes

VocalParse paper GAME technical guide (pinned)GAME ONNX guide (pinned)ROSVOT paper SOME technical guide (pinned)

Parameter guide

Auto uses GAME's official universal/unspecified language ID 0. Select a specific language only when the sung language is known.
GAME boundary-segmentation confidence threshold.
GAME boundary search radius in seconds.
Starting noise time used by GAME’s official D3PM schedule.
Number of GAME D3PM denoising steps.
GAME note-estimation confidence threshold.
After inference, Beat This final0 writes a valid fixed tempo or beat-by-beat tempo map into the MIDI while preserving every event's position in seconds. If valid beat evidence cannot be produced, the task fails instead of delivering MIDI with the model's default tempo.
ROSVOT note-boundary confidence threshold.
ROSVOT word-boundary confidence threshold.
Enable ROSVOT’s official slur consistency check.

Use cases

Turn a complete recording into editable notes, drum hits, tablature or a chord timeline.
Compare the model output with the original audio to review onsets, boundaries and source assignment.
Move MIDI and time-aligned tables into a DAW, notation editor or downstream analysis workflow.

Workflow

1Upload the complete supported file rather than a short test fragment.
2Choose the profile that matches the instrument or task, then adjust only the controls exposed by that official model.
3Review the preview against the original audio and download the result files from the completed task.

Pre-flight checklist

Do not send a dense full mix directly to a solo-instrument model; separate the target stem first when needed.
Start with the published defaults and change thresholds only when the preview shows a specific boundary or false-note problem.
Treat transcription as an editable draft: timing, note length, fingering and chord boundaries may still need musical correction.

Model evidence in this run

Models, parameters, and sources

Compare VocalParse, GAME, ROSVOT and SOME for singing-note transcription, including score-aware, accompaniment-robust and fast paths.

VocalParse 1.7B

Model

Default path for structured, score-like vocal MIDI from cleaner singing; the fixed checkpoint has no official user-tunable inference parameters.

Dedicated route: this tool uses the model path selected for this workflow.

VocalParse source (pinned)VocalParse weights (pinned)SingCrawl dataset source VocalParse paper

GAME

Model

GAME converts singing to MIDI with its official boundary, D3PM and estimation controls. Language defaults to auto/unspecified; Chinese is not forced.

Auto uses GAME's official universal/unspecified language ID 0. Select a specific language only when the sung language is known.
GAME boundary-segmentation confidence threshold.
GAME boundary search radius in seconds.
Starting noise time used by GAME’s official D3PM schedule.
Number of GAME D3PM denoising steps.
GAME note-estimation confidence threshold.
After inference, Beat This final0 writes a valid fixed tempo or beat-by-beat tempo map into the MIDI while preserving every event's position in seconds. If valid beat evidence cannot be produced, the task fails instead of delivering MIDI with the model's default tempo.

GAME source (pinned)GAME v1.0.3 weights GAME technical guide (pinned)GAME ONNX guide (pinned)

ROSVOT

Model

Robust path for singing with noise or accompaniment leakage; the current official checkpoint is trained mainly on Mandarin singing.

ROSVOT note-boundary confidence threshold.
ROSVOT word-boundary confidence threshold.
Enable ROSVOT’s official slur consistency check.

ROSVOT source (pinned)ROSVOT project page ROSVOT paper

SOME

Model

SOME is the fast, lightweight singing-to-MIDI route. It exposes no public model controls; optional MIDI tempo enhancement runs after inference.

After inference, Beat This final0 writes a valid fixed tempo or beat-by-beat tempo map into the MIDI while preserving every event's position in seconds. If valid beat evidence cannot be produced, the task fails instead of delivering MIDI with the model's default tempo.

SOME source (pinned)SOME baseline weights SOME technical guide (pinned)

Benchmark chart: VocalParse Opencpop transcription errors (paper Table 1) / ROSVOT clean and noisy robustness scores (paper Table 1)

VocalParse Opencpop transcription errors (paper Table 1)

VocalParse paper, Table 1

VocalParse, GAME, ROSVOT and SOME publish results under different datasets, languages, label rules and matching tolerances. No verified shared public protocol covers all four deployed checkpoints, so the page does not invent a combined rank.

Evaluation context	Model	MAE_pitch ↓	Detailed metrics
Ground-truth lyrics + SOFA timestamps	ROSVOTTelkNetPaper Table 1	0.38	MAE_note ↓0.45MAE_duration ↓0.40MAE_note-count ↓0.20
Audio only	VocalParse 1.7BTelkNetPaper Table 1	0.56	MAE_note ↓0.44MAE_duration ↓0.34MAE_note-count ↓0.11
Audio + ground-truth lyrics	VocalParse 1.7BTelkNetPaper Table 1	0.35	MAE_note ↓0.43MAE_duration ↓0.33MAE_note-count ↓0.11

ROSVOT clean and noisy robustness scores (paper Table 1)

ROSVOT paper, Table 1

Robust path for singing with noise or accompaniment leakage; the current official checkpoint is trained mainly on Mandarin singing.

Evaluation context	Model	COnPOff F1 (%)	Detailed metrics
Clean	ROSVOTTelkNetPaper Table 1	77.4	COn F1 (%)94.0COff F1 (%)94.5Pitch AOR (%)97.0Melody RPA (%)87.6
MUSAN noisy	ROSVOTTelkNetPaper Table 1	77.0	COn F1 (%)93.8COff F1 (%)94.4Pitch AOR (%)97.1Melody RPA (%)87.4

Benchmark audit

GAME

Parameter guide

Auto uses GAME's official universal/unspecified language ID 0. Select a specific language only when the sung language is known.
GAME boundary-segmentation confidence threshold.
GAME boundary search radius in seconds.
Starting noise time used by GAME’s official D3PM schedule.
Number of GAME D3PM denoising steps.
GAME note-estimation confidence threshold.
After inference, Beat This final0 writes a valid fixed tempo or beat-by-beat tempo map into the MIDI while preserving every event's position in seconds. If valid beat evidence cannot be produced, the task fails instead of delivering MIDI with the model's default tempo.
ROSVOT note-boundary confidence threshold.
ROSVOT word-boundary confidence threshold.
Enable ROSVOT’s official slur consistency check.

Papers / technical notes

VocalParse source (pinned)VocalParse weights (pinned)SingCrawl dataset source VocalParse paper GAME source (pinned)GAME v1.0.3 weights GAME technical guide (pinned)GAME ONNX guide (pinned)ROSVOT source (pinned)ROSVOT project page ROSVOT paper SOME source (pinned)SOME baseline weights SOME technical guide (pinned)

Demo

25時の情熱

Compare VocalParse, GAME, ROSVOT and SOME for singing-note transcription, including score-aware, accompaniment-robust and fast paths.

MIDI preview

VocalParse 1.7B

MIDI output

Under Maintenance

This tool is under maintenance. Please try again later.

Quick feedback

Write your suggestion or request here and submit it. We listen to every voice and work hard to turn good ideas into real features.

Singing Voice to MIDI · Four Models (VocalParse 1.7B) · TelkNet