fairseq S2T: Fast Speech-to-Text Modeling with fairseq

Benchmark Model Rank Results
speech-recognition-on-amifacebook/wav2vec2-conformer-rel-pos-large-960h-ft#63Word Error Rate (WER): 35.36
speech-recognition-on-amifacebook/wav2vec2-conformer-rope-large-960h-ft#64Word Error Rate (WER): 35.4
speech-recognition-on-ami-cleanedfacebook/wav2vec2-conformer-rope-large-960h-ft#64Word Error Rate (WER): 31.23
speech-recognition-on-ami-cleanedfacebook/wav2vec2-conformer-rel-pos-large-960h-ft#66Word Error Rate (WER): 31.38
speech-recognition-on-earnings-22facebook/wav2vec2-conformer-rope-large-960h-ft#65Word Error Rate (WER): 28.32
speech-recognition-on-earnings-22facebook/wav2vec2-conformer-rel-pos-large-960h-ft#66Word Error Rate (WER): 28.85
speech-recognition-on-gigaspeech-cleanedfacebook/wav2vec2-conformer-rel-pos-large-960h-ft#62Word Error Rate (WER): 15.87
speech-recognition-on-gigaspeech-cleanedfacebook/wav2vec2-conformer-rope-large-960h-ft#64Word Error Rate (WER): 16.03
speech-recognition-on-gigaspeech-testfacebook/wav2vec2-conformer-rel-pos-large-960h-ft#64Word Error Rate (WER): 16.19
speech-recognition-on-gigaspeech-testfacebook/wav2vec2-conformer-rope-large-960h-ft#66Word Error Rate (WER): 16.36
speech-recognition-on-librispeech-test-cleanfacebook/wav2vec2-conformer-rel-pos-large-960h-ft#26Word Error Rate (WER): 1.38
speech-recognition-on-librispeech-test-cleanfacebook/wav2vec2-conformer-rope-large-960h-ft#37Word Error Rate (WER): 1.53
speech-recognition-on-librispeech-test-otherfacebook/wav2vec2-conformer-rel-pos-large-960h-ft#30Word Error Rate (WER): 3.21
speech-recognition-on-librispeech-test-otherfacebook/wav2vec2-conformer-rope-large-960h-ft#41Word Error Rate (WER): 3.46
speech-recognition-on-spgispeechfacebook/wav2vec2-conformer-rel-pos-large-960h-ft#62Word Error Rate (WER): 8.37
speech-recognition-on-spgispeechfacebook/wav2vec2-conformer-rope-large-960h-ft#63Word Error Rate (WER): 8.61
speech-recognition-on-voxpopulifacebook/wav2vec2-conformer-rope-large-960h-ft#55Word Error Rate (WER): 12.07
speech-recognition-on-voxpopulifacebook/wav2vec2-conformer-rel-pos-large-960h-ft#56Word Error Rate (WER): 12.22
speech-recognition-on-voxpopuli-cleanedfacebook/wav2vec2-conformer-rope-large-960h-ft#58Word Error Rate (WER): 6.71
speech-recognition-on-voxpopuli-cleanedfacebook/wav2vec2-conformer-rel-pos-large-960h-ft#60Word Error Rate (WER): 6.88
speech-to-text-translation-on-must-c-en-deTransformer + ASR Pretrain#6Case-sensitive sacreBLEU: 22.7