LibriSpeech
Other names: tfds:librispeech
Overview
LibriSpeech is a corpus of approximately 1000 hours of read English speech with sampling rate of 16 kHz, prepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read audiobooks from the LibriVox project, and has been carefully segmented and aligned.
Description (from the source)
LibriSpeech is a corpus of approximately 1000 hours of read English speech with sampling rate of 16 kHz, prepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read audiobooks from the LibriVox project, and has been carefully segmented and aligned. It's recommended to use lazy audio decoding for faster reading and smaller dataset size: - install `tensorflow_io` library: `pip install tensorflow-io` - enable lazy decoding: `tfds.load('librispeech', builder_kwargs={'config': 'lazy_decode'})`
- Domain
- Audio & Speech
- Data type
- Audio
- Tasks
- Speech Recognition (Speech-to-Text)
- Subfields
- Speech recognition
- Languages
- ar, en, hi
- Tags
- ar, audio, en, hi, monolingual, multilingual, speech
Statistics and structure
- Number of samples
- Not available
- Number of classes
- Not available
- Number of features
- Not available
- Number of subjects / cases
- Not available
- Number of images
- Not available
- Number of files
- Not available
- Size
- Not available
- Formats
- Not available
- Missing values
- Not available
Samples note: The number of samples varies by config (5 configs in TFDS).
Licence and access
- Licence
- Unknown — no explicit licence is stated in the sources relied on
- Licence documentation
- Unknown
- Commercial use
- Unknown
- Redistribution
- Unknown
- Attribution required
- Unknown
- Access method
- Visit the source
- Access restrictions
- Not available
- Terms of use
- Not available
No explicit licence was found in the sources used; do not assume any permission before reviewing the original source.
Always review the original publisher's terms before use; the explorer grants no licence.
Source and download
- Platform
- TensorFlow Datasets catalog
- Publisher / source
- TensorFlow Datasets (TFDS)
- Identifier at the source
- tfds:librispeech
- DOI
- Not available
- Metadata source
- tfds
- Metadata last updated
- 2026-11-14
- Original page
www.openslr.org - Documentation
www.tensorflow.org - Data card
www.tensorflow.org - API
storage.googleapis.com - Metadata source
storage.googleapis.com
The explorer does not host data files; all links point to the original publisher.
Related papers
-
Paper that introduced the datasetPanayotov, Vassil, Chen, Guoguo, Povey, Daniel, Khudanpur, Sanjeev (2015). Librispeech: an ASR corpus based on public domain audio books. Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on.Listed for this dataset in the source catalogue · Evidence
Cite the dataset
Built from recorded fields only; missing fields are omitted, not guessed. It is also good practice to cite the paper that introduced the dataset, if there is one.
LibriSpeech [Data set]. (2015). TensorFlow Datasets (TFDS). http://www.openslr.org/12
[1] "LibriSpeech," TensorFlow Datasets (TFDS), 2015. [Online]. Available: http://www.openslr.org/12
@misc{librispeech2015,
title = {{LibriSpeech}},
year = {2015},
publisher = {TensorFlow Datasets (TFDS)},
url = {http://www.openslr.org/12},
howpublished = {\url{http://www.openslr.org/12}},
note = {Dataset. Accessed: 2026-11-21}
}Verification
- Verification status
- Partially verified
- Last link check
- Not checked yet
- Source verification date
- Not available
- Verification method
- Metadata imported from the official TensorFlow Datasets catalog (dataset_info.json on storage.googleapis.com and the tensorflow-datasets 4.9.10 package).
- Notes
- Counts and sizes come from the TFDS-prepared version 2.1.1. Licence not stated in TFDS metadata unless shown; check the original source. Original URL not yet checked for availability.
Source and check date of each field
| Field | Source | Date |
|---|---|---|
| original_url | TFDS dataset_info.json (location.urls) | 2026-11-14 |
| description | TFDS README.md / dataset_info.description | 2026-11-14 |
Similar datasets
Libri-Light (limited supervision subset)
Subset of Libri-light used in HuBERT for supervised fine-tuning, with 10-minute, 1-hour and 10-hour labelled subsets.
Speech Commands
An audio dataset of spoken words designed to help train and evaluate keyword spotting systems. Its primary goal is to provide a way to build and…
AccentDB
AccentDB is a multi-pairwise parallel corpus of structured and labelled accented speech. It contains speech samples from speakers of 4 non-native accents of English (8 speakers,…
TED-LIUM
The TED-LIUM corpus is English-language TED talks, with transcriptions, sampled at 16kHz. It contains about 118 hours of speech.
Common Voice
Mozilla Common Voice Dataset
XTREME-S
FLEURS is the speech version of the FLORES machine translation benchmark, covering 2000 n-way parallel sentences in n=102 languages. XTREME-S covers four task families: speech recognition,…
