List your company

Audio-visual Speech Recognition on LRS3 Object occlusion and noise

25.8WER (Babble, -10 dB)CAV2vec
24.64432.44740.2548.053Dec 16, 2025

Evaluation results

MethodLinks
CAV2vecParams=325M2025.1225.811.74.42.41.89.25.93.62.52.11.83.29.64.32.61.81.745.17.21.5
AV-RelScoreParams=437M2025.1230.114.35.32.51.710.86.84.42.82.323.710.55.32.82.11.84.55.98.41.6
AV-HuBERTParams=325M2025.1230.614.45.12.72.1117.74.332.223.910.95.232.31.84.668.61.6
AV-data2vecParams=325M2025.1232.115.15.32.5211.48.34.93.12.21.94.110.95.532.11.84.66.291.5
BRAVEnParams=673M2025.1241.122.26.12.51.814.745.630.414.95.22.219.720.38.33.52.11.87.212.218.71.8
RAVEnParams=673M2025.1243.525.48.542.816.958.942.317.95.13.125.523.311.35.43.32.59.215.2232.3
V-CAFEParams=49M2025.1254.731.614.67.35.422.745.129.316.9106.621.632.117.69.16.55.214.118.125.84.2