# LibriTTS-R: restored LibriTTS speech corpus

## Facts

- Provider: Yuma Koizumi, Heiga Zen, Shigeki Karita et al. (Google)
- License: CC-BY-4.0 (https://www.openslr.org/141/)
- Size: 198 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/mythicinfinity/libritts_r
- Access: public, readable without an account
- Category: Speech & Audio
- Source: https://arxiv.org/abs/2305.18802
- Page: https://scigantic.com/archive/b1de45fc-b281-52b0-9d2a-d495c4338f51
- Keywords: Speech & Audio, text-to-speech, Parquet, Hugging Face

## About

LibriTTS-R is a sound-quality-improved version of the LibriTTS corpus, the roughly 585-hour, 24 kHz multi-speaker English read-speech corpus built for text-to-speech research. The utterances are identical to LibriTTS; only the audio was changed, by running every sample through Miipher, a speech restoration model from Google that removes noise, reverberation and codec artefacts. The authors show that TTS models trained on LibriTTS-R produce more natural speech than the same models trained on LibriTTS, which has made it a common training set for zero-shot TTS, speech language models and vocoders. This repository is a Parquet conversion of OpenSLR SLR141: a data folder with seven split folders (dev.clean, dev.other, test.clean, test.other, train.clean.100, train.clean.360, train.other.500); train.other.500 alone holds 103 Parquet shards of about 53 GB (measured 2026-11-14; the Hub reports about 101 GB for the repo). Columns include audio, normalised and original text, speaker id, chapter id and utterance id. Load with datasets.load_dataset('mythicinfinity/libritts_r', 'clean') or read shards with pyarrow. Licensed CC BY 4.0 by the publishers on OpenSLR; cite Koizumi et al., Interspeech 2023.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("mythicinfinity/libritts_r")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/dev.clean/dev.clean-00000-of-00004.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/b1de45fc-b281-52b0-9d2a-d495c4338f51
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/b1de45fc-b281-52b0-9d2a-d495c4338f51
- Over MCP: `get_schema_card` with id `b1de45fc-b281-52b0-9d2a-d495c4338f51` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Speech & Audio datasets](https://scigantic.com/topics/speech-and-audio)
- [Text-to-speech datasets](https://scigantic.com/topics/text-to-speech)
