Skip to content

Dataset · reazon-research

Japanese speech from television broadcasts

reazon-research/reazonspeech

35,000+ hours of Japanese speech from television broadcasts for training and evaluating ASR models.

What's inside

A natural Japanese speech corpus gathered from terrestrial television broadcasts. It contains over 35,000 hours of audio; the card warns to use the data only under Article 30-4 of the Japanese Copyright Act. Format is audio files, language is Japanese.

What you can do with it

  • training a Japanese ASR model
  • fine-tuning Whisper for Japanese
  • evaluating Japanese speech recognition
Tasks
Automatic speech recognition
Languages
Japanese
Size
10M–100M rows
Downloads this month
699
Likes
139
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("reazon-research/reazonspeech")