Algorithm info: Transformer based SCNet. Analog of small version of traditional LSTM-based SCNet.
Metrics:
Metric sdr for vocals: 8.4273
Metric si_sdr for vocals: 7.4724
Metric l1_freq for vocals: 31.0975
Metric log_wmse for vocals: 11.6614
Metric aura_stft for vocals: 7.4215
Metric aura_mrstft for vocals: 8.5855
Metric bleedless for vocals: 18.7057
Metric fullness for vocals: 15.2968
Metric sdr for bass: 10.9922
Metric si_sdr for bass: 9.9405
Metric l1_freq for bass: 50.0551
Metric log_wmse for bass: 16.1790
Metric aura_stft for bass: 4.6232
Metric aura_mrstft for bass: 4.1141
Metric bleedless for bass: 19.7165
Metric fullness for bass: 18.9586
Metric sdr for drums: 10.8707
Metric si_sdr for drums: 10.2296
Metric l1_freq for drums: 34.3471
Metric log_wmse for drums: 14.6968
Metric aura_stft for drums: 7.3190
Metric aura_mrstft for drums: 6.9925
Metric bleedless for drums: 21.1681
Metric fullness for drums: 14.7368
Metric sdr for other: 5.6387
Metric si_sdr for other: 4.4018
Metric l1_freq for other: 30.2603
Metric log_wmse for other: 10.5945
Metric aura_stft for other: 5.8240
Metric aura_mrstft for other: 8.1989
Metric bleedless for other: 15.4910
Metric fullness for other: 16.9251
Date added: 2025-04-02 |
