SciganticTopicsExplore

Speech recognition datasets

25 public speech recognition datasets in the Scigantic catalog, from providers including ESPnet / CMU WAVLab, AMI Consortium; University of Edinburgh CSTR and Holy Lovenia, Samuel Cahyawijaya, Genta Indra Winata et al. . 25 can be read without an account, and 24 carry an open license. Common formats: Parquet, JSON and JSON Lines. Each opens in a hosted Jupyter notebook with the data mounted, so nothing has to be downloaded first.

DatasetProviderLicenseAccess
AMI Meeting Corpus (Kaldi-segmented ASR release)
Parquet
AMI Consortium; University of Edinburgh CSTRCC-BY-4.0public
ASCEND: spontaneous Chinese-English code-switching speech corpus
Parquet
Holy Lovenia, Samuel Cahyawijaya, Genta Indra Winata et al. CC-BY-SA-4.0public
Automatic Speech Recognition (ASR) Error Robustness
AmazonSee https://github.com/anjiefang/asr-error-robus…public
Bimodal dataset on Inner speech
Foteini LiwickiCC0public
Crowdsourced UK and Ireland English Dialect Speech (OpenSLR 83)
Parquet
Isin Demirsahin, Oddur Kjartansson, Alexander Gutkin, Clara CC-BY-SA-4.0public
Earnings-21: entity-dense earnings call ASR benchmark
JSON Lines
Miguel Del Rio, Natalie Delworth, Ryan Westerman et al. (RevCC-BY-SA-4.0public
Earnings-22: accented earnings call ASR benchmark
Parquet
Miguel Del Rio, Peter Ha, Quinten McNamara, Corey Miller, ShCC-BY-SA-4.0public
EdAcc: Edinburgh International Accents of English Corpus
Parquet
Ramon Sanabria, Nikolay Bogoychev, Nina Markl, Andrea CarmanCC-BY-SA-4.0public
FLEURS (Few-shot Learning Evaluation of Universal Representations of Speech)
TSVParquet
Googlecc-by-4.0public
FLORAS: long-form speech recognition, translation and summarization benchmark
Parquet
ESPnet / CMU WAVLabCC-BY-3.0public
Granary
JSON Lines
NVIDIAcc-by-3.0public
Industry-scale Application and Evaluation of Deep Learning for Drug Target Prediction
Noé Sturm and 18 othersCC-BY-4.0public
LibriSpeech ASR corpus
Parquet
Vassil Panayotov, Guoguo Chen, Daniel Povey, Sanjeev KhudanpCC-BY-4.0public
LibriSpeech-Long: Long-Form Speech Benchmark of Contiguous Audiobook Segments up to 4 Minutes (Google DeepMind)
Google DeepMind (Se Jin Park, Julian Salazar, Aren Jansen etCC-BY-4.0public
LoquaciousSet (LargeScaleASR): 25,000 hours of English speech
Parquet
Titouan Parcollet, Yuan Tseng, Shucong Zhang, Rogier van DalCC-BY-4.0public
Norwegian Parliamentary Speech Corpus (NPSC)
JSON
Norwegian Language Bank (Sprakbanken), National Library of NCC-BY-SA-3.0public
Omnilingual Asr Corpus
Parquet
Meta AIcc-by-4.0public
RixVox-v2: Swedish parliamentary speech dataset
Parquet
KBLab, National Library of SwedenODC-By-1.0public
The People's Speech: 30,000-hour English speech recognition corpus
Parquet
MLCommons (Daniel Galvez et al.)CC-BY-SA-4.0public
Unsupervised People's Speech
mlcommons via Source CooperativeCC-BY-4.0public
Vibravox: French speech recorded with body-conduction sensors
Parquet
Julien Hauret, Malo Olivier, Thomas Joubaud, Christophe LangCC-BY-4.0public
WAXAL: multilingual African language speech corpus (ASR and TTS)
JSONParquet
Google, with Makerere University, University of Ghana, DigitCC-BY-SA-4.0public
YODAS v3: 1.1-million-hour multilingual web speech corpus
Parquet
ESPnet / CMU WAVLabCC-BY-3.0public
YODAS2: YouTube-Oriented Dataset for Audio and Speech (long-form)
JSON
Xinjian Li, Shinnosuke Takamichi, Takaaki Saeki et al. (CMU CC-BY-3.0public
Voices Obscured in Complex Environmental Settings (VOiCES)
In-Q-TelCreative Commons BY 4.0 (see…public

Related topics