基于卷积和门控注意的两阶段视听语音增强算法

王盼蓉,贾海蓉,段淑斐

Two-stage audio-visual speech enhancement algorithm based on convolution and gated attention

Panrong WANG,Hairong JIA,Shufei DUAN

表 5 闹钟噪声5种信噪比下各模型的语音增强效果对比

Tab.5 Comparison of speech enhancement effect of various model under five signal-to-noise ratios of alarm noise

模型 PESQ SNR/dB −10 dB −5 dB 0 dB 5 dB 10 dB −10 dB −5 dB 0 dB 5 dB 10 dB Noisy 1.70 1.92 2.25 2.42 2.64 −10 −5 0 5 10 AV-ConvTasnet 2.35 2.32 2.55 2.60 2.65 4.39 4.50 4.50 4.29 4.32 MuSE 2.35 2.55 2.65 2.69 2.73 6.27 6.80 6.91 6.47 6.38 AV-Sepformer 2.67 2.81 2.86 2.92 2.93 6.77 6.79 6.82 6.62 6.76 本文模型 2.88 3.08 3.25 3.35 3.45 12.89 13.57 14.33 14.12 14.39