Dataset · formospeech
Audio and Transcript Corpus for ASR (tat_asr)
formospeech/tat_asr
Speech corpus divided into clean and noisy speech for training and evaluating ASR systems.
What's inside
The dataset includes audio recordings and text transcripts (over 100K examples total, split into train, eval, and test). The data is divided into two equally sized subsets: clean speech (reading_clean, about 51.5 hours total) and speech in challenging acoustic conditions (reading_mc, about 51.4 hours). Access is gated, and specific languages are not indicated in the fragment.
What you can do with it
- Training and fine-tuning ASR models for both clean and noisy speech recognition
- Benchmarking speech recognition robustness against acoustic noise
- Size
- 100K–1M rows
- Downloads this month
- 26.8K
- Likes
- 0
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("formospeech/tat_asr")