Research Article
Multichannel Speech Enhancement in Vehicle Environment Based on Interchannel Attention Mechanism
Table 1
Evaluation results of our proposed model compared with other methods on the same dataset. Four metrics and two SNRs are considered.
| Method | # Mics | SDR | SI-SNR | PESQ | STOI | −10 dB | −5 dB | −10 dB | −5 dB | −10 dB | −5 dB | −10 dB | −5 dB |
| Noise | 2 linear | −6.72 | 0.34 | −7.19 | 0.24 | 1.07 | 1.13 | 0.40 | 0.64 | 4 linear | −5.93 | 0.11 | −6.40 | 0.02 | 1.06 | 1.11 | 0.40 | 0.64 | 2 × 2 dB | −6.72 | 0.32 | −6.18 | 0.22 | 1.07 | 1.13 | 0.40 | 0.64 | 4 dB | −6.89 | 0.68 | −7.17 | 0.60 | 1.07 | 1.13 | 0.46 | 0.70 | +NCC | 2 linear | 7.45 | 11.98 | 6.42 | 11.30 | 1.20 | 1.76 | 0.73 | 0.87 | 4 linear | 8.89 | 12.44 | 7.86 | 11.75 | 1.28 | 1.83 | 0.75 | 0.88 | 2 × 2 dB | 7.63 | 13.14 | 7.50 | 12.37 | 1.23 | 1.85 | 0.71 | 0.89 | 4 dB | 8.38 | 14.58 | 7.60 | 13.96 | 1.30 | 2.02 | 0.76 | 0.91 | +ICD | 2 linear | 7.42 | 11.95 | 6.39 | 11.27 | 1.19 | 1.75 | 0.72 | 0.86 | 4 linear | 8.80 | 12.38 | 7.79 | 11.68 | 1.27 | 1.82 | 0.74 | 0.87 | 2 × 2 dB | 7.55 | 13.06 | 7.44 | 12.30 | 1.22 | 1.83 | 0.71 | 0.89 | 4 dB | 8.33 | 14.51 | 7.56 | 13.93 | 1.27 | 2.01 | 0.76 | 0.91 | +MCS | 2 linear | 7.40 | 11.92 | 6.36 | 11.25 | 1.19 | 1.75 | 0.72 | 0.86 | 4 linear | 8.83 | 12.40 | 7.83 | 11.72 | 1.27 | 1.82 | 0.74 | 0.87 | 2 × 2 dB | 7.52 | 13.04 | 7.42 | 12.28 | 1.21 | 1.82 | 0.70 | 0.88 | 4 dB | 8.28 | 14.46 | 7.52 | 13.79 | 1.25 | 2.00 | 0.75 | 0.90 | Proposed | 2 linear | 7.52 | 12.06 | 6.48 | 11.36 | 1.22 | 1.77 | 0.74 | 0.88 | 4 linear | 8.94 | 12.49 | 7.91 | 11.83 | 1.29 | 1.83 | 0.75 | 0.88 | 2 × 2 dB | 7.72 | 13.23 | 7.58 | 12.45 | 1.24 | 1.86 | 0.72 | 0.90 | 4 dB | 8.41 | 14.66 | 7.66 | 14.07 | 1.33 | 2.04 | 0.76 | 0.91 |
|
|
The best evaluation results are shown in bold, comparing the results of the four speech enhancement methods used in the four microphone arrays.
|