Skip to main content

Table 5 AUC results of the comparison VADs with the BLSTM model and MRCG acoustic feature on the Chinese Noisy-THCHS-30 test dataset

From: AUC optimization for deep learning-based voice activity detection

Noise type

SNR

MCE

MMSE

MaxAUCsigm

MaxAUChinge

Babble

− 10 dB

0.6276

0.6209

0.6324

0.6370

 

− 5 dB

0.7238

0.7073

0.7278

0.7362

 

0 dB

0.8165

0.7947

0.8184

0.8269

 

5 dB

0.8763

0.8586

0.8774

0.8826

 

10 dB

0.9061

0.8974

0.9080

0.9110

 

15 dB

0.9223

0.9197

0.9246

0.9280

 

20 dB

0.9358

0.9345

0.9369

0.9414

Factory

− 10 dB

0.7542

0.7479

0.7618

0.7658

 

− 5 dB

0.8355

0.8284

0.8414

0.8457

 

0 dB

0.8813

0.8761

0.8846

0.8873

 

5 dB

0.9053

0.9017

0.9075

0.9089

 

10 dB

0.9201

0.9176

0.9219

0.9231

 

15 dB

0.9314

0.9296

0.9330

0.9349

 

20 dB

0.9410

0.9390

0.9421

0.9445

Volvo

− 10 dB

0.9359

0.9352

0.9373

0.9376

 

− 5 dB

0.9472

0.9459

0.9473

0.9483

 

0 dB

0.9549

0.9529

0.9543

0.9556

 

5 dB

0.9594

0.9570

0.9584

0.9599

 

10 dB

0.9615

0.9588

0.9604

0.9621

 

15 dB

0.9620

0.9590

0.9607

0.9628

 

20 dB

0.9616

0.9584

0.9601

0.9628