Skip to content

Dataset · LEMAS-Project

150k hours of speech across 10 languages

LEMAS-Project/LEMAS-Dataset-train

A large training corpus and evaluation set with word-level alignment for ASR and TTS across 10 languages.

What's inside

A dataset from the LEMAS-Project: a training set of 150k+ hours and an evaluation set of 500 utterances per language. It covers 10 languages; the metadata lists Italian, Portuguese, Spanish, French, German, and Vietnamese. Word-level alignment is included, audio is stored in MP3, and keys contain the utterance ID and file path.

What you can do with it

  • Training multilingual ASR models
  • TTS fine-tuning across 10 languages
  • Evaluating word-level alignment
Tasks
Text → speech, Automatic speech recognition
Languages
Italian, Portuguese, Spanish, French, German, Vietnamese, Indonesian, Russian and 2 more
Size
100M–1B rows
Downloads this month
16.1K
Likes
91

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("LEMAS-Project/LEMAS-Dataset-train")