Skip to main content

Table 1 PESQ comparison on the test set at different input SNRs of unseen noise environments

From: Speech enhancement from fused features based on deep neural network and gated recurrent unit network

Noise

SNR (dB)

Unprocessed

DNN

CNN

LSTM

GRU

DNN-GRU

Pink

20

3.059

3.330

2.478

3.253

3.181

3.528

15

2.710

3.075

2.956

3.0151

2.951

3.314

10

2.358

2.792

2.144

2.724

2.664

2.963

5

1.962

2.491

2.228

2.411

2.324

2.790

0

1.606

2.118

1.826

2.049

1.940

2.462

− 5

1.292

1.708

1.371

1.675

1.515

2.114

White

20

2.791

3.156

2.381

3.101

3.080

3.271

15

2.443

2.921

2.922

2.889

2.847

2.966

10

2.102

2.659

1.953

2.618

2.605

2.837

5

1.720

2.356

2.145

2.345

2.287

2.674

0

1.424

2.057

1.693

2.026

1.957

2.386

− 5

1.200

1.613

1.270

1.671

1.597

1.975

Battle

20

3.152

3.362

2.550

3.339

3.241

3.353

15

2.831

3.160

2.974

3.103

3.009

3.253

10

2.503

2.880

2.265

2.818

2.723

2.880

5

2.126

2.549

2.188

2.522

2.406

2.655

0

1.796

2.205

1.844

2.156

2.062

2.332

− 5

1.471

1.829

1.446

1.739

1.693

2.175

Factory

20

3.240

3.436

3.396

3.346

3.269

3.642

15

2.907

3.240

2.970

3.146

3.059

3.423

10

2.572

2.992

2.289

2.879

2.810

3.229

5

2.039

2.563

2.209

2.463

2.368

2.937

0

1.683

2.221

1.811

2.107

1.983

2.592

− 5

1.370

1.797

1.400

1.725

1.593

2.285

F16

20

3.117

3.416

2.527

3.368

3.253

3.587

15

2.429

2.943

3.013

3.145

3.137

3.308

10

2.440

2.958

2.265

2.850

2.757

3.169

5

2.065

2.676

2.292

2.557

2.429

2.882

0

1.726

2.321

1.906

2.182

2.066

2.538

− 5

1.424

1.965

1.491

1.844

1.744

2.419

Destroy

20

3.191

3.346

2.546

3.375

3.260

3.602

15

2.878

2.223

2.967

3.174

3.047

3.440

10

2.547

2.971

2.290

2.898

2.774

3.276

5

2.193

2.738

2.273

2.635

2.487

2.950

0

1.828

2.391

1.933

2.205

2.128

2.535

− 5

1.468

1.995

1.533

1.773

1.708

2.393