Algorithm info:
chunk_size = 485100
overlap = 2
batch_size= 2
using colab inference script (https://github.com/jarredou/Music-Source-Separation-Training-Colab-Inference/)
https://huggingface.co/pcunwa/Kim-Mel-Band-Roformer-FT/tree/main
-jarredou
Metrics:
Metric sdr for instrum: 17.3594
Metric si_sdr for instrum: 17.2582
Metric l1_freq for instrum: 39.9223
Metric log_wmse for instrum: 14.1540
Metric aura_stft for instrum: 13.4003
Metric aura_mrstft for instrum: 16.6453
Metric bleedless for instrum: 46.3084
Metric fullness for instrum: 27.7811
Metric sdr for vocals: 11.0519
Metric si_sdr for vocals: 10.6013
Metric l1_freq for vocals: 38.9968
Metric log_wmse for vocals: 14.1540
Metric aura_stft for vocals: 8.1673
Metric aura_mrstft for vocals: 10.0257
Metric bleedless for vocals: 39.3007
Metric fullness for vocals: 15.7731
Date added: 2025-01-30 |