Hi-Fi Multi-Speaker English TTS Dataset (Hi-Fi TTS)
- Provider
- Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg, Yang Zhang (NVIDIA)
- License
- CC-BY-4.0 (source)
- Size
- 95 files
- Format
- Parquet (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/MikhailT/hifi-tts)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Speech & Audio
- Source
- https://arxiv.org/abs/2104.01497
About Hi-Fi Multi-Speaker English TTS Dataset (Hi-Fi TTS)
Hi-Fi TTS is a multi-speaker English dataset for training text-to-speech models, built by NVIDIA from LibriVox public domain audiobooks and Project Gutenberg texts. It contains about 291.6 hours of speech from 10 speakers, each with at least 17 hours, sampled at 44.1 kHz, which makes it one of the few open corpora with enough high-sample-rate audio per speaker to train studio-quality single- and multi-speaker TTS and neural vocoders. The original release sorts each speaker's audio into 'clean' and 'other' quality bins. This repository is a Parquet conversion of the OpenSLR SLR109 release: a data folder of 93 Parquet shards named by split and quality (for example dev.clean, dev.other, test.clean, train.clean), each row holding the audio and its text (measured 2026-11-14; the Hub reports about 166 GB). Open with datasets.load_dataset('MikhailT/hifi-tts') or read the shards with pyarrow or pandas. The dataset is released by its authors on OpenSLR under CC BY 4.0; cite Bakhturina et al., 'Hi-Fi Multi-Speaker English TTS Dataset' (arXiv 2104.01497).
How to load Hi-Fi Multi-Speaker English TTS Dataset (Hi-Fi TTS) in Python
from datasets import load_dataset
ds = load_dataset("MikhailT/hifi-tts")
print(ds)Data structure
File types in sampled files: Parquet
Files
Example file: data/dev.clean-00000-of-00001-2202c1655cccc3db.parquet
Analyze this dataset without downloading it
On Scigantic, Hi-Fi Multi-Speaker English TTS Dataset (Hi-Fi TTS) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Speech & Audio datasets
- LJ Speech single-speaker English TTS dataset
- YODAS2: YouTube-Oriented Dataset for Audio and Speech (long-form)
- LibriTTS-R: restored LibriTTS speech corpus
- LibriTTS multi-speaker English text-to-speech corpus
- LibriSpeech-Long: Long-Form Speech Benchmark of Contiguous Audiobook Segments up to 4 Minutes (Google DeepMind)
- CML-TTS multilingual speech synthesis dataset
- Omnilingual Asr Corpus
- VoiceBank+DEMAND noisy speech database (16 kHz)