Common Voice: A Massively-Multilingual Speech Corpus
By Rosana Ardila, Megan Branson, Kelly Davis, Michael Henretty, M. Kohler, Josh Meyer, Reuben Morais, Lindsay Saunders, Francis M. Tyers, Gregor Weber. 2019
2,502 citations recorded by the source corpus.
The Common Voice corpus is a massively-multilingual collection of transcribed speech intended for speech technology research and development. Common Voice is designed for Automatic Speech Recognition purposes but can be useful in other domains (e.g. language identification). To achieve scale and sustainability, the Common Voice project employs crowdsourcing for both data collection and data validation. The most recent release includes 29 languages, and as of November 2019 there are a total of 38 languages collecting data. Over 50,000 individuals have participated so far, resulting in 2,500 hours of collected audio. To our knowledge this is the largest audio corpus in the public domain for speech recognition, both in terms of number of hours and number of languages. As an example use case for Common Voice, we present speech recognition experiments using Mozilla’s DeepSpeech Speech-to-Text toolkit. By applying transfer learning from a source English model, we find an average Character Error Rate improvement of 5.99 ± 5.48 for twelve target languages (German, French, Italian, Turkish, Catalan, Slovenian, Welsh, Irish, Breton, Tatar, Chuvash, and Kabyle). For most of these languages, these are the first ever published results on end-to-end Automatic Speech Recognition.
Subfields: Artificial Intelligence, Speech & Audio, Signal Processing
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 Speech & Audio
- Jinyu Li
- Wei-Ning Hsu
- Yu Zhang
- Gabriel Synnaeve
- Abdel-rahman Mohamed
- Shinji Watanabe
- Alexei Baevski
- Yossi Adi
In Signal Processing
- Jinyu Li
- Yu Zhang
- Shinji Watanabe
- Wei-Ning Hsu
- Dong Yu
- Tara N. Sainath
- Abdel-rahman Mohamed
- Gabriel Synnaeve
Citations in the corpus
- FLEURS: FEW-Shot Learning Evaluation of Universal Representations of Speech (2022, methodology): “Such recent progress has been made possible with the release of both large-scale pre-training and evaluation datasets like Multilingual LibriSpeech [13], VoxPopuli [14], CoVoST2 [15], CommonVoice [16], and the re-use of existing datasets like BABEL [17].”
- MAESTRO: Matched Speech Text Representations through Modality Matching (2022, methodology): “Multilingual ASR: Following [11], we use 429k hours of public unlabeled speech corpora: VoxPopuli [33], CommonVoice [34], MLS [35] and BABEL [36] to pre-train multilingual ASR models.”
- mSLAM: Massively multilingual joint pre-training for speech and text (2022, background): “CoVoST 2 Speech Translation: CoVoST 2 (Wang et al., 2021b) is a multilingual speech translation (ST) dataset created by professional translation of the Common Voice speech corpus (Ardila et al., 2020).”
- XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale (2021, methodology): “We conduct experiments on CoVoST-2 [28], a multilingual speech translation benchmark based on CommonVoice [21].”
- The People's Speech: A Large-Scale Diverse English Speech Recognition Dataset for Commercial Usage (2021, background): “Unlike Common Voice, Our dataset uses existing audio and transcripts from resources on the web, which are licensed appropriately for commercial usage by their creators (rather than soliciting volunteers to curate the dataset [15]).”
- BigSSL: Exploring the Frontier of Large-Scale Semi-Supervised Learning for Automatic Speech Recognition (2021, methodology): “The inference results on the test sets are scored via corresponding Kaldi scripts, while for the Common Voice set, we take the extra step of dropping punctuation before evaluating the word error rates.”
- Unsupervised Speech Recognition (2021, background): “This is a multilingual corpus of read speech for 38 languages (Ardila et al., 2020).”
- SpeechStew: Simply Mix All Available Speech Recognition Data to Train One Large Neural Network (2021, neutral): “Common Voice [32].”
- 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): “Out-of-domain pre-training We examine the out-of-domain pre-training setting using the Common Voice (CV) ASR corpus (Ardila et al., 2020).”
- MLS: A Large-Scale Multilingual Dataset for Speech Research (2020, background): “(ii) CommonVoice [4], a more scalable solution, with more than 30 languages available, which keeps growing with 4500 (validated) hours currently available.”
- High Fidelity Neural Audio Compression (2022, background): “The ground truth is 16bits 24kHz wave. speech segments from DNS Challenge 4 (Dubey et al., 2022) and the Common Voice dataset (Ardila et al., 2019).”
- Audiobox: Unified Audio Generation with Natural Language Prompts (2023, methodology): “First, we annotate a set of 1,946 recordings sampled from diverse sources, including LibriTTS (Zen et al., 2019), Common Voice (Ardila et al., 2019), Switchboard (Godfrey et al., 1992), Fisher (Cieri, Christopher, et al. , 2004,2005a,,), Spotify (Clifton et al., 2020), AudioSet (Gemmeke et al., 2017”
- Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale (2023, methodology): “Prior work uses labels to make the mapping between input (text and audio style) and output (speech) more deterministic to reduce underfitting [60, 46].”
- AudioPaLM: A Large Language Model That Can Speak and Listen (2023, positive): “• Common Voice [Ardila et al., 2020] consists of text paired with recordings where people were asked to read the text aloud.”
- High-Fidelity Audio Compression with Improved RVQGAN (2023, methodology): “For speech, we use the DAPS dataset [26], the clean speech segments from DNS Challenge 4 [10], the Common Voice dataset [2], and the VCTK dataset [40].”
- YourTTS: Towards Zero-Shot Multi-Speaker TTS and Zero-Shot Voice Conversion for everyone (2021, positive): “Nevertheless, to show the potential of our model for adaptation to new speakers/recording conditions, we selected samples from 20 to 61 seconds of speech for 2 Portuguese and 2 English speakers (1M/1F) in the Common Voice [38] dataset.”