Datasets:

lamooon
/

vocal-v1

onset_sec float64 2.3 284	offset_sec float64 2.72 284	pitch_midi int64 43 83	pitch_hz float64 98 988	source_dataset stringclasses 1 value
15.029199	15.295866	60	261.63	mir_st500
15.379167	15.775	60	261.63	mir_st500
15.800033	16.159408	60	261.63	mir_st500
16.329167	16.671875	60	261.63	mir_st500
16.7	17.055208	64	329.63	mir_st500
17.219792	17.439583	65	349.23	mir_st500
17.619792	18.367708	60	261.63	mir_st500
18.5	20	58	233.08	mir_st500
21.119792	21.567708	57	220	mir_st500
21.579167	22.015625	58	233.08	mir_st500
23.35	23.775	58	233.08	mir_st500
23.783333	24.185417	60	261.63	mir_st500
24.25	24.600033	55	196	mir_st500
24.629167	25.439583	58	233.08	mir_st500
25.51	25.983333	57	220	mir_st500
29.119792	29.471875	57	220	mir_st500
29.51	29.944792	57	220	mir_st500
29.969792	30.303125	62	293.66	mir_st500
30.439583	30.751042	62	293.66	mir_st500
30.879167	31.2	64	329.63	mir_st500
31.329199	31.551074	64	329.63	mir_st500
31.73	32.063542	65	349.23	mir_st500
32.209375	32.991667	57	220	mir_st500
33.1875	33.490625	55	196	mir_st500
35.25	35.600033	55	196	mir_st500
35.739583	36.063542	57	220	mir_st500
36.079167	36.447917	55	196	mir_st500
36.5	36.850033	55	196	mir_st500
36.89375	37.243783	53	174.61	mir_st500
37.275	37.625033	55	196	mir_st500
38.88125	39.231283	53	174.61	mir_st500
39.25	39.68125	57	220	mir_st500
39.68125	40.11	55	196	mir_st500
40.11	40.559375	55	196	mir_st500
40.566667	40.855208	53	174.61	mir_st500
40.979167	41.759375	60	261.63	mir_st500
43.219792	43.615625	60	261.63	mir_st500
43.619792	44	60	261.63	mir_st500
44.109375	44.447917	60	261.63	mir_st500
44.469792	44.959408	60	261.63	mir_st500
44.979167	45.247917	64	329.63	mir_st500
45.4	45.663542	65	349.23	mir_st500
45.879167	46.43125	60	261.63	mir_st500
46.75	47.815625	58	233.08	mir_st500
49.25	49.434375	55	196	mir_st500
49.456283	49.706283	57	220	mir_st500
49.75	50.271908	58	233.08	mir_st500
51.61	51.90625	58	233.08	mir_st500
52.019792	52.319792	60	261.63	mir_st500
52.419792	52.767708	55	196	mir_st500
52.97	53.695833	57	220	mir_st500
57.409375	57.759375	57	220	mir_st500
57.785417	58.209375	57	220	mir_st500
58.209375	58.623958	62	293.66	mir_st500
58.65	59.103125	62	293.66	mir_st500
59.109375	59.519792	64	329.63	mir_st500
59.539583	59.907292	64	329.63	mir_st500
59.95	60.317708	65	349.23	mir_st500
60.39	61.279199	69	440	mir_st500
61.309408	61.695866	67	392	mir_st500
62.229199	62.655241	62	293.66	mir_st500
63.529167	63.807292	65	349.23	mir_st500
63.97	64.671875	69	440	mir_st500
64.819792	65.439583	67	392	mir_st500
65.689583	66.463542	61	277.18	mir_st500
66.97	67.263542	65	349.23	mir_st500
67.47	70.079167	67	392	mir_st500
70.95	71.327083	57	220	mir_st500
71.53	71.871875	60	261.63	mir_st500
71.879167	72.255208	64	329.63	mir_st500
72.279167	72.703125	65	349.23	mir_st500
72.819792	73.823958	67	392	mir_st500
74.13	74.527083	69	440	mir_st500
74.55	74.94375	69	440	mir_st500
75.83	76.063542	65	349.23	mir_st500
76.33	77.439583	67	392	mir_st500
77.619792	78.015625	69	440	mir_st500
78.119792	78.879167	69	440	mir_st500
79.389583	79.807292	72	523.25	mir_st500
79.835417	80.261458	72	523.25	mir_st500
80.2875	80.655208	70	466.16	mir_st500
80.75	81.161458	69	440	mir_st500
81.20625	81.535417	65	349.23	mir_st500
81.65	81.983333	65	349.23	mir_st500
82.089583	82.463542	67	392	mir_st500
82.469792	82.911458	69	440	mir_st500
82.939583	83.327083	70	466.16	mir_st500
83.4	83.967708	67	392	mir_st500
85.215625	85.6	57	220	mir_st500
85.629167	85.983333	60	261.63	mir_st500
86	86.367708	64	329.63	mir_st500
86.379167	86.879167	65	349.23	mir_st500
86.909375	88.063542	67	392	mir_st500
88.209375	88.607292	69	440	mir_st500
88.7	89.439583	69	440	mir_st500
89.939583	90.4	65	349.23	mir_st500
90.429167	91.583333	67	392	mir_st500
91.709375	92.063542	72	523.25	mir_st500
92.15	92.927083	69	440	mir_st500
93.489583	93.919792	72	523.25	mir_st500

End of preview. Expand in Data Studio

Vocal Melody Transcription Dataset v1

Training data for a monophonic vocal melody transcription model. The model uses a ROSVOT-style architecture (MERT encoder → U-Net w/ Conformer bottleneck → onset/pitch/frame heads) and is designed to receive Demucs v4-separated vocal audio at inference time.

Datasets

Source	Tracks	Description
MIR-ST500	385	Pop songs with manual onset/offset/pitch annotations
DALI	~4,927	Large-scale vocal annotations aligned to audio (10 batch tars)
MedleyDB	107	Multitrack recordings with Melody2 F0→note converted annotations
Total	~5,420	Matched audio+label pairs across all sources

All audio is resampled to 24kHz mono WAV, peak-normalized to -1dB. Labels are unified CSV format: onset_sec, offset_sec, pitch_midi, pitch_hz, source_dataset.

Files on this repo

File	Size	Contents
`vocal_v1.tar`	~1.3GB	MIR-ST500 processed audio + labels
`vocal_v1_dali_batch{1-10}.tar`	~4GB each	DALI processed audio + labels (10 batches)
`vocal_v1_medleydb.tar`	~915MB	MedleyDB processed audio (24kHz) + note-level labels
`MedleyDB_v1.tar`	~8.8GB	Raw MedleyDB V1: 122 MIX wavs + 116 vocal stems (not used directly in training pipeline)
`oneshots.tar`	~972MB	1,204 curated vocal oneshots for vocal bleed augmentation
`vocal_v1_augmented.tar`	—	OBSOLETE (on-the-fly augmentation used instead)

Augmentation (on-the-fly)

Applied during training only (never to val/test):

Pitch shift: ±4 semitones (label-aware — adjusts pitch annotations)
Time stretch: 0.85x–1.15x
Noise injection: SNR 10–40dB
Vocal bleed: overlay random oneshots at SNR 25–40dB
Downsample-resample: through 16kHz/22.05kHz
Random EQ: 3-band, ±3dB gain

Training Pipeline

run.sh (SLURM) auto-downloads all tars (skipping augmented and raw MedleyDB_v1), extracts to data/, rebuilds train/val/test splits stratified by source (seed 42), then trains with on-the-fly augmentation.

Label Format

onset_sec, offset_sec, pitch_midi, pitch_hz, source_dataset
0.52, 0.89, 60.0, 261.63, mir_st500

Pitch targets use bin offset: bin 0 = unvoiced, bin 1 = MIDI 21 (A0), so pitch_bin = midi_note - 21 + 1.

Downloads last month: 166

Size of the auto-converted Parquet files:

389 kB

Number of rows:

21,304