wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations

Benchmark Model Rank Results
speech-recognition-on-amifacebook/wav2vec2-large-960h-lv60-self#56Word Error Rate (WER): 29.45
speech-recognition-on-amifacebook/wav2vec2-large-960h#65Word Error Rate (WER): 35.72
speech-recognition-on-amifacebook/wav2vec2-base-960h#66Word Error Rate (WER): 38.98
speech-recognition-on-ami-cleanedfacebook/wav2vec2-large-960h-lv60-self#58Word Error Rate (WER): 25.83
speech-recognition-on-ami-cleanedfacebook/wav2vec2-large-960h#67Word Error Rate (WER): 32.13
speech-recognition-on-ami-cleanedfacebook/wav2vec2-base-960h#68Word Error Rate (WER): 35.41
speech-recognition-on-earnings-22facebook/wav2vec2-large-960h-lv60-self#59Word Error Rate (WER): 23.05
speech-recognition-on-earnings-22facebook/wav2vec2-large-960h#67Word Error Rate (WER): 36.28
speech-recognition-on-earnings-22facebook/wav2vec2-base-960h#69Word Error Rate (WER): 41.41
speech-recognition-on-gigaspeech-cleanedfacebook/wav2vec2-large-960h-lv60-self#58Word Error Rate (WER): 14.76
speech-recognition-on-gigaspeech-cleanedfacebook/wav2vec2-large-960h#67Word Error Rate (WER): 19.09
speech-recognition-on-gigaspeech-cleanedfacebook/wav2vec2-base-960h#69Word Error Rate (WER): 22.3
speech-recognition-on-gigaspeech-testfacebook/wav2vec2-large-960h-lv60-self#61Word Error Rate (WER): 15.01
speech-recognition-on-gigaspeech-testfacebook/wav2vec2-large-960h#70Word Error Rate (WER): 19.37
speech-recognition-on-gigaspeech-testfacebook/wav2vec2-base-960h#72Word Error Rate (WER): 22.58
speech-recognition-on-librispeech-test-cleanfacebook/wav2vec2-large-960h-lv60-self#23Word Error Rate (WER): 1.36
speech-recognition-on-librispeech-test-cleanwav2vec 2.0 with Libri-Light#54Word Error Rate (WER): 1.8
speech-recognition-on-librispeech-test-cleanfacebook/wav2vec2-base-960h#97Word Error Rate (WER): 2.76
speech-recognition-on-librispeech-test-cleanfacebook/wav2vec2-large-960h#103Word Error Rate (WER): 3.03
speech-recognition-on-librispeech-test-otherwav2vec 2.0 with Libri-Light#24Word Error Rate (WER): 3.0
speech-recognition-on-librispeech-test-otherfacebook/wav2vec2-large-960h-lv60-self#33Word Error Rate (WER): 3.33
speech-recognition-on-librispeech-test-otherwav2vec 2.0#61Word Error Rate (WER): 4.1
speech-recognition-on-librispeech-test-otherfacebook/wav2vec2-large-960h#97Word Error Rate (WER): 6.53
speech-recognition-on-librispeech-test-otherfacebook/wav2vec2-base-960h#105Word Error Rate (WER): 7.95
speech-recognition-on-spgispeechfacebook/wav2vec2-large-960h-lv60-self#60Word Error Rate (WER): 7.81
speech-recognition-on-spgispeechfacebook/wav2vec2-large-960h#68Word Error Rate (WER): 13.31
speech-recognition-on-spgispeechfacebook/wav2vec2-base-960h#70Word Error Rate (WER): 18.44
speech-recognition-on-timitwav2vec 2.0#1Percentage error: 8.3
speech-recognition-on-voxpopulifacebook/wav2vec2-large-960h-lv60-self#61Word Error Rate (WER): 13.88
speech-recognition-on-voxpopulifacebook/wav2vec2-large-960h#66Word Error Rate (WER): 21.2
speech-recognition-on-voxpopulifacebook/wav2vec2-base-960h#67Word Error Rate (WER): 24.16
speech-recognition-on-voxpopuli-cleanedfacebook/wav2vec2-large-960h-lv60-self#56Word Error Rate (WER): 6.27
speech-recognition-on-voxpopuli-cleanedfacebook/wav2vec2-large-960h#68Word Error Rate (WER): 12.87
speech-recognition-on-voxpopuli-cleanedfacebook/wav2vec2-base-960h#69Word Error Rate (WER): 15.16