2024:Polyphonic Transcription Results

Submissions

Group	Extended Abstract	Methods Used	Summary Onset F1	Holistic Note F1
wlazbzfll	PDF	CRNN + regression onset&offset	0.7066	0.1607
teamWLY	PDF	FiLM with CNN + LSTM + regression onset & offset	0.9592	0.8465
Transkun V2 (Baseline 1)	PDF	ViT + Neural SemiCRF	0.9490	0.8764
Transkun V2 Aug (Baseline 2)	PDF	ViT + Neural SemiCRF + Data Augmentation	0.9648	0.9081
hFT-Transformer	PDF	Two stacks of Transformers for Frequency and Time + regression onset&offset	0.9416	0.8359

• average note onset F1 = average of note onset F1 on all three datasets
• holistic note F1 = average of note onset+offset+velocity on Maestro and SMD

Detailed Results on Maestro

	Activation			Note Onset			Note Onset+Offset			Note Onset+Offset+ vel.
Team	prec	recall	F1	prec	recall	F1	prec	recall	F1	prec	recall	F1
wlazbzfll	0.7324	0.7673	0.7445	0.7411	0.6511	0.6845	0.3537	0.3158	0.3294	0.1527	0.1353	0.1415
teamWLY	0.953	0.9261	0.939	0.9944	0.9719	0.9829	0.9277	0.907	0.9171	0.9061	0.886	0.8958
Transkun V2	0.9576	0.9489	0.953	0.9956	0.9714	0.9832	0.9465	0.9238	0.9349	0.9411	0.9186	0.9296
Transkun V2 Aug	0.9495	0.9522	0.9505	0.9971	0.9715	0.984	0.9437	0.9197	0.9314	0.9386	0.9149	0.9264
hFT-Transformer	0.9537	0.9108	0.9307	0.9964	0.9545	0.9745	0.9244	0.8861	0.9045	0.9144	0.8768	0.8948

	pedal activation			pedal onset			pedal onset+offset
Team	prec	recall	F1	prec	recall	F1	prec	recall	F1
wlazbzfll	0.9429	0.9432	0.9419	0.7775	0.7812	0.7785	0.7377	0.7405	0.7383
teamWLY	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
Transkun V2	0.9671	0.9453	0.9541	0.8909	0.8421	0.8642	0.8632	0.8165	0.8377
Transkun V2 Aug	0.9546	0.9416	0.9454	0.8883	0.8116	0.8453	0.8497	0.7790	0.8102
hFT-Transformer	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A

• offset derivations on MAPS deviates strongly from a Normal distribution, suggesting potential annotation issues
• N/A on pedals means that no pedal is included in the transcribed results
• Aug means the model is trained with data augmentation

Detailed Results on MAPS

	Activation			Note Onset			Note Onset+Offset			Note Onset+Offset+ vel.
Team	prec	recall	F1	prec	recall	F1	prec	recall	F1	prec	recall	F1
wlazbzfll	0.8021	0.7338	0.7626	0.741	0.6822	0.7024	0.4421	0.4166	0.4243	0.1793	0.1698	0.1724
teamWLY	0.9049	0.8394	0.8691	0.918	0.9008	0.9089	0.6896	0.6772	0.683	0.4653	0.4571	0.461
Transkun V2	0.887	0.8252	0.8535	0.8671	0.9048	0.8849	0.6325	0.6613	0.6461	0.4351	0.4551	0.4446
Transkun V2 Aug	0.9446	0.8334	0.8843	0.9396	0.9056	0.9219	0.7105	0.6854	0.6975	0.5596	0.5401	0.5495
hFT-Transformer	0.916	0.716	0.8019	0.8747	0.8841	0.8789	0.5689	0.5756	0.5719	0.3592	0.3636	0.3612

	pedal activation			pedal onset			pedal onset+offset
Team	prec	recall	F1	prec	recall	F1	prec	recall	F1
wlazbzfll	0.8565	0.8217	0.833	0.5164	0.582	0.539	0.3301	0.378	0.3477
teamWLY	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
Transkun V2	0.8498	0.8547	0.8449	0.6521	0.7182	0.6732	0.4903	0.5427	0.5088
Transkun V2 Aug	0.8893	0.8389	0.8583	0.7313	0.7529	0.7343	0.5499	0.5650	0.5532
hFT-Transformer	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A

• offset derivations on MAPS deviates strongly from a Normal distribution, suggesting potential annotation issues
• N/A on pedals means that no pedal is included in the transcribed results
• Aug means the model is trained with data augmentation

Detailed Results on SMD

	Activation			Note Onset			Note Onset+Offset			Note Onset+Offset+ vel.
Team	prec	recall	F1	prec	recall	F1	prec	recall	F1	prec	recall	F1
wlazbzfll	0.7766	0.7862	0.7775	0.751	0.7323	0.7329	0.3915	0.3958	0.3895	0.1704	0.1696	0.1682
teamWLY	0.9237	0.9357	0.9289	0.9945	0.9771	0.9857	0.9005	0.8849	0.8926	0.8042	0.7902	0.7971
Transkun V2	0.9203	0.9491	0.934	0.9816	0.9766	0.979	0.9013	0.8968	0.899	0.8255	0.8211	0.8232
Transkun V2 Aug	0.9389	0.9518	0.9448	0.997	0.9801	0.9884	0.9284	0.9128	0.9205	0.8974	0.8823	0.8897
hFT-Transformer	0.9291	0.8968	0.9115	0.9875	0.9563	0.9713	0.8834	0.856	0.8692	0.7897	0.7651	0.7769

	pedal activation			pedal onset			pedal onset+offset
Team	prec	recall	F1	prec	recall	F1	prec	recall	F1
wlazbzfll	0.9402	0.9459	0.9415	0.7921	0.8018	0.7956	0.727	0.735	0.7299
teamWLY	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
Transkun V2	0.9364	0.9507	0.942	0.8722	0.8101	0.8388	0.803	0.7471	0.773
Transkun V2 Aug	0.9491	0.9428	0.9447	0.8788	0.8043	0.8383	0.8208	0.7526	0.7837
hFT-Transformer	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A

• offset derivations on MAPS deviates strongly from a Normal distribution, suggesting potential annotation issues
• N/A on pedals means that no pedal is included in the transcribed results
• Aug means the model is trained with data augmentation