Skip to content

Dataset · MushanW

Cleaned English audio corpus based on Common Voice 21.0

MushanW/GLOBE_V3

Filtered Common Voice recordings with 24 kHz sample rate and perfect transcription match by Whisper V3 Large.

What's inside

Containing between 100k and 1 million recordings, the dataset is built on the English part of Common Voice 21.0. It includes only raw audio files sampled at 24 kHz whose text transcripts match Whisper V3 Large recognition 100% (CER == 0), along with original speaker age, gender, and accent metadata.

What you can do with it

  • Training high-quality text-to-speech (TTS) models with accent, gender, and age control
  • Training and evaluating automatic speech recognition (ASR) and voice classification models
Tasks
Text → audio, Automatic speech recognition, Audio → audio, Audio classification
Languages
English
Size
100K–1M rows
Downloads this month
2.4K
Likes
4

How to load it

from datasets import load_dataset

ds = load_dataset("MushanW/GLOBE_V3")