Skip to content

Dataset · MLCommons

Unsupervised People's Speech

MLCommons/unsupervised_peoples_speech

Over one million hours of English audio from Archive.org for ASR and audio classification.

What's inside

An English audio dataset collected from Archive.org files, containing over one million hours of speech from diverse speakers. CC-BY and CC-BY-SA licenses permit academic and commercial use. Provided as audio files without annotations.

What you can do with it

  • Train or fine-tune an ASR model on a large English corpus
  • Pre-train an unannotated audio encoder or audio classifier
  • Test ASR robustness across diverse speakers and recordings
Tasks
Automatic speech recognition, Audio classification
Languages
English
Downloads this month
21.3K
Likes
83

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("MLCommons/unsupervised_peoples_speech")