Libri-Light: A Benchmark For Asr With Limited Or No Supervision

This paper introduces a new corpus of English speech suitable for training speech recognition systems under limited or no supervision. It is derived from open source audio books in the LibriVox project and governmental speech recordings and contain over 70K hours of speech, making it, to our knowledge, the largest corpus of speech freely available. We also provide baseline systems and evaluation metrics working under three settings: (1) the zero resource/unsupervised setting (ABX), (2) the semi-supervised setting (PER, CER) and (3) the distant supervision setting (WER). Settings (2) and (3) use limited textual resources (10 minutes to 10 hours) aligned with the speech. Setting (3) uses large amounts of unaligned text. They are evaluated on the standard LibriSpeech dev and test set for comparison with the supervised state-of-the-art.