Skip to content

Dataset · thanhnew2001

103 hours of annotated Vietnamese speech for ASR

thanhnew2001/VietSuperSpeech

Audio dataset with Vietnamese transcriptions (32,267 fragments, 16 kHz) for speech recognition.

What's inside

A corpus for Vietnamese speech recognition with a total volume of 103.18 hours, split into training (29,041 records) and validation (3,226 records) sets. Audio has a 16,000 Hz sample rate and an average fragment length of about 12 seconds. The compilation combines three source datasets (nguoivietdailynews, nguyenkhangofficial, trinhlieu) and is formatted in Icefall (train.json and dev.json).

What you can do with it

  • Training or fine-tuning ASR models (e.g., Whisper, Zipformer) for the Vietnamese language
  • Evaluating Vietnamese speech recognition quality on the provided dev split
Size
10K–100K rows
Downloads this month
20.2K
Likes
6

How to load it

from datasets import load_dataset

ds = load_dataset("thanhnew2001/VietSuperSpeech")