Robust Speech Recognition via Large-Scale Weak Supervision

Benchmark Model Rank Results
speech-recognition-on-amiopenai/whisper-large-v3#31Word Error Rate (WER): 14.86
speech-recognition-on-amiopenai/whisper-large-v3-turbo#34Word Error Rate (WER): 15.16
speech-recognition-on-ami-cleanedopenai/whisper-large-v3#38Word Error Rate (WER): 13.63
speech-recognition-on-ami-cleanedopenai/whisper-large-v3-turbo#42Word Error Rate (WER): 13.87
speech-recognition-on-common-voice-frenchWhisper (Large v2)#9Test WER: 13.9%
speech-recognition-on-common-voice-germanWhisper (Large v2)#7Test WER: 6.4%
speech-recognition-on-common-voice-spanishWhisper (Large v2)#3Test WER: 5.6%
speech-recognition-on-earnings-22openai/whisper-large-v3-turbo#22Word Error Rate (WER): 11.07
speech-recognition-on-earnings-22openai/whisper-large-v3#28Word Error Rate (WER): 11.59
speech-recognition-on-gigaspeech-cleanedopenai/whisper-large-v3#21Word Error Rate (WER): 8.35
speech-recognition-on-gigaspeech-cleanedopenai/whisper-large-v3-turbo#25Word Error Rate (WER): 8.47
speech-recognition-on-gigaspeech-testopenai/whisper-large-v3#20Word Error Rate (WER): 8.4
speech-recognition-on-gigaspeech-testopenai/whisper-large-v3-turbo#22Word Error Rate (WER): 8.52
speech-recognition-on-librispeech-test-cleanopenai/whisper-large-v3#39Word Error Rate (WER): 1.55
speech-recognition-on-librispeech-test-cleanopenai/whisper-large-v3-turbo#81Word Error Rate (WER): 2.13
speech-recognition-on-librispeech-test-otheropenai/whisper-large-v3#43Word Error Rate (WER): 3.52
speech-recognition-on-librispeech-test-otheropenai/whisper-large-v3-turbo#49Word Error Rate (WER): 3.7
speech-recognition-on-spgispeechopenai/whisper-large-v3#29Word Error Rate (WER): 2.71
speech-recognition-on-spgispeechopenai/whisper-large-v3-turbo#32Word Error Rate (WER): 2.79
speech-recognition-on-voxpopuliopenai/whisper-large-v3#49Word Error Rate (WER): 8.7
speech-recognition-on-voxpopuliopenai/whisper-large-v3-turbo#54Word Error Rate (WER): 11.22
speech-recognition-on-voxpopuli-cleanedopenai/whisper-large-v3#51Word Error Rate (WER): 4.53
speech-recognition-on-voxpopuli-cleanedopenai/whisper-large-v3-turbo#61Word Error Rate (WER): 7.02
speech-to-speech-translation-on-fleurs-x-engWhisperV2#6ASR-BLEU: 23.5