Dataset · Flux9665
Multilingual eBible TTS phrases
Flux9665/BibleMMS
Audio and text for text-to-speech: 2,000 phrases per language, over 7,000 languages.
What's inside
A dataset for speech synthesis: audio and text generated for 2,000 utterances per language. Texts are taken from freely licensed subsets of eBible, and the audio was generated by MMS TTS models. Languages with ISO-639-3 from the card are covered; over 7,000 languages in total. The material is related to the paper Meta Learning Text-to-Speech Synthesis in over 7000 Languages (Interspeech 2024).
What you can do with it
- Train a multilingual TTS model
- Evaluate TTS transfer to low-resource languages
- Fine-tune LoRA for speech synthesis in specific languages
- Tasks
- Text → speech
- Size
- 100K–1M rows
- Downloads this month
- 553
- Likes
- 83
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("Flux9665/BibleMMS")