LibriSpeech
Benchmarks
TaskDatasetSOTA resultTrendResults
LibriSpeech (LS) clean
1.67WER
21
LibriSpeech (test-other)
10.3WER
21
LibriSpeech clean (test)
4.73WER
21
LibriSpeech Clean 100 source and target longer than 1 min
0.78FAD
20
LibriSpeech English (test)
0.717Speaker Similarity
20
LibriSpeech (test-clean)
0.59BWER
20
LibriSpeech 960h
4.52WER
20
LibriSpeech Case 2: Source and Target < 1 min (test)
0.23WER
19
LibriSpeech Case 4: source duration > 1 min, target duration < 1 min
41Word Error Rate
19
LibriSpeech
1.1WER
19
LibriSpeech 1000 (test-clean)
5.29WER
19
LibriSpeech Out-of-Vocabulary queries train-clean-100
82MAP
18
LibriSpeech train-clean-100 (In-Vocabulary queries)
86MAP
18
Librispeech
100DSR
18
LibriSpeech
0Normalized RMSE
18
LibriSpeech (test)
1.84WER
18
LibriSpeech (dev)
1.8WER
18
LibriSpeech
3.771PESQ
18
LibriSpeech clean PC (test)
1.98WER (%)
17
LibriSpeech clean (test)
4.64PESQ
16
LibriSpeech-PC (test-clean)
1.93WER
16
LibriSpeech 30% packet loss (test)
4.09OVRL
16
LibriSpeech clean (test)
1.9WERH
16
LibriSpeech clean-100 1.0 (train)
0.83Unigram Consistency
15
LibriSpeech Gaussian noise, Avg
15.6WER
15