Music Source Separation On Musdb18

评估指标

SDR (avg)

SDR (bass)

SDR (drums)

SDR (other)

SDR (vocals)

评测结果

各个模型在此基准测试上的表现结果

						Paper Title	Repository
Sparse HT Demucs (fine tuned)	9.20	10.47	10.83	6.41	9.37	Hybrid Transformers for Music Source Separation
Hybrid Transformer Demucs (f.t.)	9.00	9.78	10.08	6.42	9.20	Hybrid Transformers for Music Source Separation
Band-Split RNN (semi-sup.)	8.97	8.16	10.15	7.08	10.47	Music Source Separation with Band-split RNN
TFC-TDF-UNet (v3)	8.34	8.45	8.44	6.86	9.59	Sound Demixing Challenge 2023 Music Demixing Track Technical Report: TFC-TDF-UNet v3
Band-Split RNN	8.23	7.51	8.58	6.62	10.21	Music Source Separation with Band-split RNN
Hybrid Demucs	7.72	8.67	8.58	5.59	8.04	Hybrid Spectrogram and Waveform Source Separation
KUIELab-MDX-Net	7.54	7.86	7.33	5.95	9.00	KUIELab-MDX-Net: A Two-Stream Neural Network for Music Demixing
CDE-HTCN	6.89	7.92	7.33	4.92	7.37	Hierarchic Temporal Convolutional Network With Cross-Domain Encoder for Music Source Separation	-
Attentive-MultiResUNet	6.81	5.88	7.63	5.14	8.57	An Efficient Short-Time Discrete Cosine Transform and Attentive MultiResUNet Framework for Music Source Separation	-
DEMUCS (extra)	6.79	7.60	7.58	4.69	7.29	Music Source Separation in the Waveform Domain
CWS-PResUNet	6.77	5.93	6.38	5.84	8.92	CWS-PResUNet: Music Source Separation with Channel-wise Subband Phase-aware ResUNet
D3Net	6.68	6.20	7.36	5.37	7.80	D3Net: Densely connected multidilated DenseNet for music source separation
DLMNet	6.47	7.29	7.05	4.62	6.91	Contrastive Learning based Deep Latent Masking for Music Source Separation	-
Conv-TasNet (extra)	6.32	7.00	7.11	-	6.74	Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation
UMXL	6.316	6.015	7.148	4.889	7.213	Open-Unmix - A Reference Implementation for Music Source Separation	-
DEMUCS	6.28	7.01	6.86	4.42	6.84	Music Source Separation in the Waveform Domain
TAK2	6.04	5.40	6.81	4.80	7.16	MMDenseLSTM: An efficient combination of convolutional and recurrent neural networks for audio source separation
D3Net	6.01	5.25	7.01	4.53	7.24	D3Net: Densely connected multidilated DenseNet for music source separation
Spleeter (MWF)	5.91	5.51	6.71	4.02	6.86	Spleeter: A Fast And State-of-the Art Music Source Separation Tool With Pre-trained Models	-
LaSAFT+GPoCM	5.88	5.63	5.68	4.87	7.33	LaSAFT: Latent Source Attentive Frequency Transformation for Conditioned Source Separation

0 of 27 row(s) selected.