SpeechStew: Simply Mix All Available Speech Recognition Data to Train One Large Neural Network

Benchmark Model Rank Results
speech-recognition-on-chime-6-dev-gss12SpeechStew (1B)Word Error Rate (WER): 31.9
speech-recognition-on-librispeech-test-cleanSpeechStew (100M)Word Error Rate (WER): 2.0
speech-recognition-on-librispeech-test-cleanSpeechStew (1B)Word Error Rate (WER): 1.7
speech-recognition-on-librispeech-test-otherSpeechStew (100M)Word Error Rate (WER): 4.0
speech-recognition-on-librispeech-test-otherSpeechStew (1B)Word Error Rate (WER): 3.3
speech-recognition-on-tedliumSpeechStew (100M)Word Error Rate (WER): 5.3
speech-recognition-on-wsj-eval92Speechstew 100MWord Error Rate (WER): 1.3