Skip to main content
EURASIP Journal on Advances in Signal Processing

Table 3 Average PESQ results among five models under mismatch input SNRs

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

SNR

Unprocessed

DNN

CNN

LSTM

GRU

DNN-GRU

17 dB

3.032

3.172

3.069

3.292

3.266

3.587

8 dB

2.405

2.568

2.574

2.778

2.682

2.893

2 dB

1.973

2.156

1.987

2.329

2.237

2.489

− 7 dB

1.418

1.725

1.525

1.883

1.863

1.928

Average

2.207

2.405

2.289

2.571

2.512

2.774

Back to article page