LibriTTS-R English Speech Corpus for Text-to-Speech

LibriTTS-R is a multi-speaker English speech corpus containing approximately 585 hours of read speech at a 24kHz sampling rate. The dataset is a sound quality improved version of the original LibriTTS corpus published in 2019. It was adapted for the Hugging Face datasets library by the user 'mythicinfinity'.

Use Cases

• Training text-to-speech models using paired audio and text transcriptions.
• Developing multi-speaker voice cloning systems using the speaker identities and corresponding audio samples.
• Improving speech synthesis quality by leveraging the 24kHz high-fidelity audio samples for better sound generation.
• Benchmarking speech synthesis models on a large-scale, read English speech corpus.

Strengths

• Contains approximately 585 hours of speech data.
• Audio is sampled at a high 24kHz rate for improved sound quality.
• Includes multiple speaker identities for multi-speaker modeling.

Limitations

• Specific speaker count, row count, and detailed column information are unknown.
• The dataset's original collection methodology and potential biases are not detailed.
• The license for use and redistribution is unspecified.

Provenance

Source

Adapted from the LibriTTS corpus hosted on OpenSLR (http://www.openslr.org/60/).

Collection Method

Derived from read English audiobooks; the specific recording and processing method for the 'R' version is not detailed.

Time Range

null

Freshness

The dataset card was last updated on 2024-02-09, but the underlying corpus was published in 2019.

Geography

Primarily US English speakers, based on the original LibriTTS corpus description.