TD-SpeakerBeam

2024年3月15日修改

Part Conf

sample_rate:16k

mix_type:clean_double

mix_src:2

train data set :Libri2Mix from librispeech-clean-100 (13900)

test data set :Libri2Mix from librispeech-test-clean (3000)

task:sep_clean

segment: 3

segment_aux: 3

Final Metrics

"si_sdr": 13.527310012269766,

"si_sdr_imp": 13.526672833841847,

"sdr": 14.195532584777578,

"sdr_imp": 14.105653583237528,

"sir": Infinity,

"sir_imp": NaN,

"sar": 14.195532584777578,

"sar_imp": 14.105653583237528,

"stoi": 0.9112053981698651,

"stoi_imp": 0.19811220804088683

In paper, SDR is 11.17 for the wsjmix dataset and 17.24 for the csjmix dataset.

Example

input_si_sdr:1.1229279041290283,

input_sdr:1.17094655782011,

input_sir:inf,

input_sar:1.17094655782011,

input_stoi:0.7189430854389222,

si_sdr:14.631566047668457,

sdr:14.833218857494998,

sir:inf,

sar:14.833218857494998,

stoi:0.97248967906343,

mix

61-70968-0000_8455-210777-0012

TD-SpeakerBeam​