LRS3
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
LRS3 mixed with VGGSound noises (test)
3.25PESQ
18
LRS3 Noisy w.r.t Missing Modality 0.3 Masking Ratio (test)
5.77WER (%)
16
LRS3 Noisy w.r.t Missing Modality 0.1 Masking Ratio (test)
4.07Word Error Rate (%)
16
LRS3 Noisy w.r.t Viewpoint 15° Offset (test)
4.05WER
16
LRS3 Noisy w.r.t Viewpoint 5° Offset (test)
3.86WER
16
LRS3 High-Resource 433h labelled v1 (test)
0.012WER
16
LRS3 high-resource (test)
23.1WER
16
LRS3 Low-Resource 30h labelled v1 (test)
2.3WER (%)
15
LRS3
0.008WER
14
LRS3-2Mix (test)
17.5SDRi
11
LRS3 high-resource (test)
1.3WER
11
LRS3
2.29WER
10
LRS3 (test)
43Overall Score
10
LRS3-T Fr-En
97.04BLEU
10
LRS3-T Es-En
97.04BLEU
10
LRS3 NoiseX babble noise (test)
11.2WER (SNR -5dB)
9
LRS3 MUSAN babble noise (test)
10.15WER (SNR -5dB)
9
LRS3 (test)
87Accuracy
9
LRS3 v0.4 (test)
2.3WER
9
LRS3 Samples with > 100 Frames (test)
2.6WER (10 dB)
8
LRS3 noisy
4.2Average Error Rate
8
LRS3 2Mix (test)
17.5SI-SNRi
8
LRS3 433 h 0 dB SNR
1.7WER
7
LRS3 433 h 5 dB SNR
1.3WER
7
LRS3 Pixelated face
26WER (Babble, -10 dB)
7