GigaSpeech

10,000-hour multi-domain English ASR corpus spanning audiobooks, podcasts, and YouTube.

• Task: asr
• Languages: en
• Hours: 10000
• Domain: audiobook, podcast, youtube
• License: see source terms
• Homepage: https://github.com/SpeechColab/GigaSpeech

Recommendation

Choose when you need scale and domain diversity beyond LibriSpeech (podcasts, YouTube). Requires accepting terms and an access request, and is large — plan storage carefully. No VoxKitchen recipe; download manually then point a dir-ingest ASR pipeline at your local copy.

Getting the data

Obtain from the dataset homepage.

Request access via the GigaSpeech GitHub repository and follow the download instructions there. The XL split (10,000 h) requires ~1 TB of storage; the M/L splits are more manageable for prototyping.

Suggested processing

A recommended VoxKitchen pipeline ships in the repository at voxkitchen/templates/pipelines/asr-training-data.yaml — run it with vkit docker run.