Loading...
Convert isolated instrument recordings into MIDI for notation and arranging. Choose a model for your source and review its supported instruments.
Input
Supports inputs in formats such as mp3 / wav / flac / ogg / m4a / aac / aif / aiff / mp4 / mov / mkv
Supported formats
Output
MIDI
Use case
Edit notes, instrumentation and performance details.
Choose by input and output needs, then comparable tests, published evaluations and capabilities. Missing scores do not imply lower quality; more parameters or a newer release do not prove superiority. This is a suggested usage order, not an overall leaderboard.
Cost, time, formats and parameters depend on the selected model.
Route clean or separated solo guitar, jazz saxophone, jazz double bass, solo violin, or monophonic melody recordings to the matching specialist model.
GAPS converts clean or separated solo guitar into note events and MIDI, without string positions, frets, continuous F0 or Pitch Bend. Adjust MIDI velocity by comparing with the original recording.
Guitar-FL converts guitar solos from the François Leduc data domain into discrete note events and MIDI; FL does not mean frame-level. It does not output continuous F0 or Pitch Bend.
FiloSax targets monophonic jazz saxophone solos and exports discrete pitch, onset, and offset events as MIDI. It does not output continuous pitch, slide, or vibrato tracks.
FiloBass targets monophonic jazz double bass (upright bass), not general or electric bass. Use a clean or separated stem; kick and low-frequency bleed can create false notes.
VioPTT outputs violin note events and one of four technique labels—détaché, flageolet, spiccato, pizzicato—or none. MIDI velocity is fixed, not measured performance dynamics.
STRAdi Online targets clean solo violin with fully causal inference and no future context. It exports discrete MIDI plus onset, offset and pitch CSV; the public repository does not ship a score-following system.
STRAdi Offline is the matched non-causal, full-context reference route. It exports the same discrete MIDI and onset, offset and pitch CSV; it is not a published score-following system either.
SwiftF0 0.2.0 uses its official pitch tracking and note segmentation for monophonic recordings. Outputs MIDI, frame-level pitch CSV and a plot.
Convert isolated instrument, vocal or drum tracks to MIDI using the matching official model. Vocal harmony also supports the v1.6 checkpoint. Instrument label refinement remains experimental.
Spotify Basic Pitch transcribes polyphonic performances from nearly any instrument, including voice, into MIDI and optional note-event CSV with pitch bends. It works best on one instrument at a time.
Scores apply only to the cited dataset, metric and model version. Leading this table does not mean a current world record or a result from this service. Capabilities and tests can justify prioritizing models without comparable scores.
Scores come from the cited public sources; blue rows are models available in this tool.
The sources use different training or task settings, so compare numbers only within the stated protocol.
Public sources checked: 2026-09-10
| Test setting | Model | F1 (%) | Detailed metrics |
|---|---|---|---|
| URMP violin tracks · Table 1 | VioPTT w/ augmentation (paper)VioPTT paper Table 1 | 84.50 | Precision (%)86.10Recall (%)83.60 |
| Bach10 violin tracks · Table 1 | VioPTT w/ augmentation (paper)VioPTT paper Table 1 | 69.90 | Precision (%)68.10Recall (%)71.80 |
This update changes runtime dependencies for GAPS, Guitar-FL, FiloSax and FiloBass, with unchanged weights. Paper results retain their datasets. FiloBass Table 5 evaluates other algorithms on that dataset; the current FiloBass checkpoint has no published score.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 10/10 rows · 3 metrics
| Rank | Test setting | Model | Note onset F1 (%) | Precision (%) | Recall (%) |
|---|---|---|---|---|---|
| #1 | GuitarSet · zero-shot · 50 ms | GAPS (Maman initialization) | 88.10 | 91.10 | 85.90 |
| #2 | GuitarSet · zero-shot · 50 ms | Guitar-FL | 87.30 | 88.00 | 87.10 |
| #3 | GuitarSet · zero-shot · 50 ms | GAPS (Kong initialization) | 87.00 | 91.60 | 83.70 |
| #4 | GuitarSet · zero-shot · 50 ms | GAPS (scratch) | 86.10 | 92.40 | 81.80 |
| #5 | GuitarSet · zero-shot · 50 ms | Maman (MusicNet EM {}_{\textit{EM}} ) [ 1 ] | 82.90 | 86.60 | 80.40 |
| #6 | GuitarSet · zero-shot · 50 ms | Maman (Guitar) [ 1 ] | 82.20 | 86.70 | 79.70 |
| #7 | GuitarSet · zero-shot · 50 ms | Zang et al. [ 8 ] (Synthtab) | 70.20 | — | — |
| #8 | GuitarSet · zero-shot · 50 ms | Kong et al. [ 10 ] | 54.80 | 67.50 | 49.70 |
| #9 | GuitarSet · zero-shot · 50 ms | Kong et al. (w/ aug) | 50.30 | 80.60 | 44.00 |
| #10 | GuitarSet · zero-shot · 50 ms | MT3 [ 16 ] | 32.00 | — | — |
This update changes runtime dependencies for GAPS, Guitar-FL, FiloSax and FiloBass, with unchanged weights. Paper results retain their datasets. FiloBass Table 5 evaluates other algorithms on that dataset; the current FiloBass checkpoint has no published score.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 6/6 rows · 3 metrics
| Rank | Test setting | Model | Note onset F1 (%) | Precision (%) | Recall (%) |
|---|---|---|---|---|---|
| #1 | FrançoisLeduc test · 50 ms | GAPS (Maman initialization) | 84.80 | 85.50 | 84.20 |
| #2 | FrançoisLeduc test · 50 ms | Guitar-FL | 84.70 | 83.90 | 85.50 |
| #3 | FrançoisLeduc test · 50 ms | Lu et al. [ 14 ] | 80.00 | 83.60 | 77.30 |
| #4 | FrançoisLeduc test · 50 ms | Omnizart [ 17 ] | 67.10 | 63.00 | 72.10 |
| #5 | FrançoisLeduc test · 50 ms | Basic Pitch [ 15 ] | 66.10 | 54.60 | 85.00 |
| #6 | FrançoisLeduc test · 50 ms | MT3 [ 16 ] | 52.40 | 48.80 | 57.00 |
This update changes runtime dependencies for GAPS, Guitar-FL, FiloSax and FiloBass, with unchanged weights. Paper results retain their datasets. FiloBass Table 5 evaluates other algorithms on that dataset; the current FiloBass checkpoint has no published score.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 4/4 rows · 3 metrics
| Rank | Test setting | Model | Note onset F1 (%) | Precision (%) | Recall (%) |
|---|---|---|---|---|---|
| #1 | GAPS test · 50 ms | GAPS (Maman initialization) | 94.30 | 95.00 | 93.60 |
| #2 | GAPS test · 50 ms | GAPS (Kong initialization) | 94.00 | 94.60 | 93.40 |
| #3 | GAPS test · 50 ms | GAPS (scratch) | 93.40 | 94.90 | 92.10 |
| #4 | GAPS test · 50 ms | Guitar-FL | 92.10 | 92.90 | 91.40 |
This update changes runtime dependencies for GAPS, Guitar-FL, FiloSax and FiloBass, with unchanged weights. Paper results retain their datasets. FiloBass Table 5 evaluates other algorithms on that dataset; the current FiloBass checkpoint has no published score.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 3/3 rows · 3 metrics
| Rank | Test setting | Model | Note onset F1 (%) | Precision (%) | Recall (%) |
|---|---|---|---|---|---|
| #1 | FiloSax · 25-track test · 50 ms | FiloSax | 96.19 | 96.47 | 95.90 |
| #2 | FiloSax · 25-track test · 50 ms | Başaran et al. | 90.01 | 87.68 | 92.57 |
| #3 | FiloSax · 25-track test · 50 ms | CREPE Notes | 82.67 | 82.35 | 83.05 |
This update changes runtime dependencies for GAPS, Guitar-FL, FiloSax and FiloBass, with unchanged weights. Paper results retain their datasets. FiloBass Table 5 evaluates other algorithms on that dataset; the current FiloBass checkpoint has no published score.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 3/3 rows · 3 metrics
| Rank | Test setting | Model | Note onset F1 (%) | Precision (%) | Recall (%) |
|---|---|---|---|---|---|
| #1 | Omnibook · 50 ms | FiloSax | 75.43 | 74.70 | 76.96 |
| #2 | Omnibook · 50 ms | Başaran et al. | 73.68 | 70.60 | 77.98 |
| #3 | Omnibook · 50 ms | CREPE Notes | 70.41 | 70.93 | 70.63 |
This update changes runtime dependencies for GAPS, Guitar-FL, FiloSax and FiloBass, with unchanged weights. Paper results retain their datasets. FiloBass Table 5 evaluates other algorithms on that dataset; the current FiloBass checkpoint has no published score.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
± indicates standard deviation.
Showing 3/3 rows · 4 metrics
| Rank | Test setting | Model | Note onset F1 (%) | Recall (%) | Precision (%) | Overlap (%) |
|---|---|---|---|---|---|---|
| #1 | FiloBass dataset · 50 ms | Melodyne | 78.95 ± 15.02 | 79.52 ± 14.77 | 78.48 ± 15.41 | 87.94 ± 3.91 |
| #2 | FiloBass dataset · 50 ms | CREPE Notes | 72.89 ± 12.68 | 74.11 ± 12.09 | 71.81 ± 13.33 | 78.77 ± 2.68 |
| #3 | FiloBass dataset · 50 ms | Basic Pitch | 62.73 ± 5.55 | 81.28 ± 6.26 | 51.40 ± 6.28 | 65.24 ± 4.51 |
SwiftF0 0.2.0 uses the author's latest pitch-benchmark: 10 corpora, 8 scored conditions and 50 cents tolerance. This is pitch F1; the difference from RMVPE is statistically undetermined. Asymmetric 95% CIs are retained. REAPER is unranked due to incomplete results; the older 0.1.x paper does not score 0.2.0.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 18/18 rows · 4 metrics
| Rank | Test setting | Model | Pitch F1@50c | Voicing F1 | CI95 lo | CI95 hi |
|---|---|---|---|---|---|---|
| #1 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | SwiftF0 0.2.0TelkNet | 0.778 | 0.839 | 0.765 | 0.792 |
| #2 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | RMVPE | 0.768 | 0.837 | 0.752 | 0.783 |
| #3 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | FCPE | 0.728 | 0.812 | 0.712 | 0.742 |
| #4 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | TorchCREPE | 0.691 | 0.761 | 0.673 | 0.706 |
| #5 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | CREPE | 0.689 | 0.781 | 0.672 | 0.704 |
| #6 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | PESTO | 0.680 | 0.850 | 0.663 | 0.697 |
| #7 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | SHS | 0.657 | 0.773 | 0.640 | 0.672 |
| #8 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | Praat | 0.651 | 0.815 | 0.634 | 0.667 |
| #9 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | RAPT | 0.640 | 0.799 | 0.621 | 0.655 |
| #10 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | HarmoF0 | 0.639 | 0.727 | 0.622 | 0.653 |
| #11 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | SWIPE | 0.610 | 0.761 | 0.589 | 0.624 |
| #12 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | SPICE | 0.602 | 0.751 | 0.582 | 0.619 |
| #13 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | Harvest | 0.600 | 0.782 | 0.584 | 0.614 |
| #14 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | YAAPT | 0.560 | 0.842 | 0.539 | 0.580 |
| #14 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | PENN | 0.560 | 0.720 | 0.539 | 0.579 |
| #14 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | DIO | 0.560 | 0.710 | 0.541 | 0.577 |
| #17 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | Basic Pitch | 0.557 | 0.775 | 0.539 | 0.573 |
| #18 | pitch-benchmark · 10 corpora · 8 altered conditions · 50 cents | pYIN | 0.506 | 0.656 | 0.482 | 0.525 |
Tsumugi Other and Vocal Harmony report COnP, not COnF1. Published results cover v1.5; default Vocal Harmony v1.6 has no published score. Instrument refinement remains experimental, with datasets shown separately.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 2/2 rows · 1 metrics
| Rank | Test setting | Model | COnP (%) |
|---|---|---|---|
| #1 | In-house real audio | other_v1_5 | 77.01 |
| #2 | In-house real audio | other | 73.18 |
Tsumugi Other and Vocal Harmony report COnP, not COnF1. Published results cover v1.5; default Vocal Harmony v1.6 has no published score. Instrument refinement remains experimental, with datasets shown separately.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 2/2 rows · 1 metrics
| Rank | Test setting | Model | COnP (%) |
|---|---|---|---|
| #1 | Held-out MIR-ST500 | vocal_harmony_v1_5 | 68.14 |
| #2 | Held-out MIR-ST500 | vocal_harmony | 60.52 |
Tsumugi Other and Vocal Harmony report COnP, not COnF1. Published results cover v1.5; default Vocal Harmony v1.6 has no published score. Instrument refinement remains experimental, with datasets shown separately.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 2/2 rows · 2 metrics
| Rank | Test setting | Model | DrumKit F1 (%) | AllPercussion F1 (%) |
|---|---|---|---|---|
| #1 | Audio-aligned · 50 ms · canonical GM | drums_v1_5 | 68.90 | 40.44 |
| #2 | Audio-aligned · 50 ms · canonical GM | drums | 61.57 | 18.79 |
Tsumugi Other and Vocal Harmony report COnP, not COnF1. Published results cover v1.5; default Vocal Harmony v1.6 has no published score. Instrument refinement remains experimental, with datasets shown separately.
Ranks are computed within each table and evaluation context, retaining ties. Change the metric to reorder; scroll horizontally for every column. Missing values are unranked, never zero.
Public sources checked: 2026-09-23
Showing 2/2 rows · 1 metrics
| Rank | Test setting | Model | ACC (%) |
|---|---|---|---|
| #1 | Held-out RWC-I | Instrument refinement model | 74.50 |
| #2 | Held-out RWC-I | Base instrument classifier | 71.30 |
Scores come from the linked base checkpoint, not a new measurement of the model running here.
Public sources checked: 2026-09-10
| Dataset | Model | F_no-offset | Detailed metrics |
|---|---|---|---|
| Molina | 0.52 | Frame accuracy ↑0.63F_with-offset ↑0.35 | |
| GuitarSet | 0.79 | Frame accuracy ↑0.70F_with-offset ↑0.56 | |
| MAESTRO | 0.71 | Frame accuracy ↑0.38F_with-offset ↑0.11 | |
| Slakh | 0.42 | Frame accuracy ↑0.44F_with-offset ↑0.21 | |
| PHENICX | 0.49 | Frame accuracy ↑0.53F_with-offset ↑0.35 |
ISMIR - 2024
GAPS converts clean or separated solo guitar into note events and MIDI, without string positions, frets, continuous F0 or Pitch Bend. Adjust MIDI velocity by comparing with the original recording.
ICASSP - 2024
Guitar-FL converts guitar solos from the François Leduc data domain into discrete note events and MIDI; FL does not mean frame-level. It does not output continuous F0 or Pitch Bend.
SMC - 2024
FiloSax targets monophonic jazz saxophone solos and exports discrete pitch, onset, and offset events as MIDI. It does not output continuous pitch, slide, or vibrato tracks.
ISMIR - 2023
FiloBass targets monophonic jazz double bass (upright bass), not general or electric bass. Use a clean or separated stem; kick and low-frequency bleed can create false notes.
ICASSP - 2026
VioPTT outputs violin note events and one of four technique labels—détaché, flageolet, spiccato, pizzicato—or none. MIDI velocity is fixed, not measured performance dynamics.
arXiv - 2025
SwiftF0 0.2.0 uses its official pitch tracking and note segmentation for monophonic recordings. Outputs MIDI, frame-level pitch CSV and a plot.
Official technical guide - 2026
Convert isolated instrument, vocal or drum tracks to MIDI using the matching official model. Vocal harmony also supports the v1.6 checkpoint. Instrument label refinement remains experimental.
ICASSP - 2022
Spotify Basic Pitch transcribes polyphonic performances from nearly any instrument, including voice, into MIDI and optional note-event CSV with pitch bends. It works best on one instrument at a time.
Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.