Libri-Light: A Benchmark for ASR with Limited or No Supervision

By Jacob Kahn, Morgane Rivière, Weiyi Zheng, Evgeny Kharitonov, Qiantong Xu, Pierre-Emmanuel Mazar'e, Pierre-Emmanuel Mazaré, Julien Karadayi, Vitaliy Liptchinsky, Ronan Collobert and 6 more. 2019

820 citations recorded by the source corpus.

We introduce a new collection of spoken English audio suitable for training speech recognition systems under limited or no supervision. It is derived from open-source audio books from the LibriVox project. It contains over 60K hours of audio, which is, to our knowledge, the largest freely-available corpus of speech. The audio has been segmented using voice activity detection and is tagged with SNR, speaker ID and genre descriptions. Additionally, we 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 sets for comparison with the supervised state-of-the-art.

Subfields: Artificial Intelligence, Natural Language Processing, Speech & Audio

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Speech & Audio

Citations in the corpus

Self-supervised Pretraining of Visual Features in the Wild (2021, background): “0 [2] is learned on 53 K hours of audio [28].”
wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations (2020, methodology): “For the latter we follow the preprocessing of [27] resulting in 53.”
MAESTRO: Matched Speech Text Representations through Modality Matching (2022, methodology): “Monolingual ASR: We use unlabeled English-only speech from the LibriLight corpus [29] to pre-train monolingual models.”
XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale (2021, methodology): “…pretrained models (XLSR-53, VP-100K) as well as baselines with English-only pretraining: XMEF JT, the best performing setup of [30] for En → X directions as well as wav2vec 2.0 pre-trained on 60K hours of English Libri-light data and fine-tuned following the same protocol as XLS-R [33, 7].”
The People's Speech: A Large-Scale Diverse English Speech Recognition Dataset for Commercial Usage (2021, background): “This work is a first step towards creating a speech dataset that covers all speakers and environments (not just English audiobooks [22, 25, 26]) for open use [27].”
Unispeech-Sat: Universal Speech Representation Learning With Speaker Aware Pre-Training (2021, neutral): “Previous works use Librispeech [20] or Librivox [16] datasets for pre-training, which limits the pre-training model since the input data are all extracted from the audiobook.”
BigSSL: Exploring the Frontier of Large-Scale Semi-Supervised Learning for Automatic Speech Recognition (2021, background): “As a baseline, we have listed SoTA word error rates for each task inside SpeechStew, and recorded the performance of the ConformerXXL RNN-T network pretrained with Libri-Light data trained on SpeechStew [30].”
HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units (2021, methodology): “When the HuBERT model is pre-trained on either the standard Librispeech 960h [24] or the Libri-Light 60k hours [25], it either matches or improves upon the state-of-theart wav2vec 2.”
SpeechStew: Simply Mix All Available Speech Recognition Data to Train One Large Neural Network (2021, neutral): “LibriLight [47] is approximately 60k hours of unlabelled data, in-domain with Lib-riSpeech.”
Robust wav2vec 2.0: Analyzing Domain Shift in Self-Supervised Pre-Training (2021, methodology): “The six domains are: (1) LibriSpeech [26] ( LS ) and Libri-light [27] ( LL ), which contain about 960 and 60K hours of 16kHz crowd-sourced audiobook recordings, respectively, derived from the LibriVox project; (2) TED-LIUM v3 [28] ( TD ), which contains 452 hours of TED conference audio in 16kHz; (3”
VoxPopuli: A Large-Scale Multilingual Speech Corpus for Representation Learning, Semi-Supervised Learning and Interpretation (2021, methodology): “Speech-to-Text and Speech-to-Speech Translation Apart from machine translation (Koehn, 2005), the European Parliament open data has fostered the development of corpora for speech-totext translation and for simultaneous interpretation.”
Self-Training and Pre-Training are Complementary for Speech Recognition (2020, positive): “For the latter we follow the pre-processing of Kahn et al. (2020; [33]) resulting in 53.2k hours of audio.”
Pushing the Limits of Semi-Supervised Learning for Automatic Speech Recognition (2020, methodology): “0 pre-training [6] with utterances from the "unlab-60k" subset of Libri-Light [2].”
Improved Noisy Student Training for Automatic Speech Recognition (2020, methodology): “The gradational methods prove to be beneﬁcial for the low-supervised-performance task LibriSpeech 100-860, while it does not show much effect for the high-supervised-performance task LibriSpeech-LibriLight.”
Iterative Pseudo-Labeling for Speech Recognition (2020, neutral): “Audio data for our experiments comes from two sources: LIBRISPEECH, containing 960h of audio and paired transcriptions, and audio from LIBRIVOX (54K hours of audio) extracted following [7].”
Unsupervised Pretraining Transfers Well Across Languages (2020, methodology): “Here, we present results on the CPC features trained on the recently released Libri-light 60K dataset[29].”
Learning Robust and Multilingual Speech Representations (2020, neutral): “Data augmentation (Park et al., 2019), self-training (Synnaeve et al., 2019; Kahn et al., 2019a) have advanced the state-of-the-art performance on English speech recognition.”