Dataset · speechcolab
10,000 hours of English speech with transcripts
speechcolab/gigaspeech
Audiobooks, podcasts, and YouTube with transcripts for training English ASR and TTS.
What's inside
GigaSpeech is an English speech corpus with 10,000 hours of labeled audio for supervised training. Transcripts are gathered from audiobooks, podcasts, and YouTube, covering reading and spontaneous speech on various topics. Data is in audio and text, in English.
What you can do with it
- Train or fine-tune an ASR model like Whisper
- Check recognition quality across different domains
- Use audio and texts for TTS experiments
- Tasks
- Automatic speech recognition, Text → speech, Text → audio
- Languages
- English
- Size
- 10M–100M rows
- Downloads this month
- 13.7K
- Likes
- 178
- Access
- on request on Hugging Face
Models trained on it
- biodatlab/ThonburianTTSText → speech 8.3K
- fixie-ai/ultravox-v0_2Audio text → text 1.8K
- 11mlabs/indri-0.1-124m-ttsText → speech 982
- mradermacher/tiny-audio-GGUF 434
- mradermacher/Open-LiveTranslate-v0-En-Zh-GGUF 431
- oruk/oruk-loop-153MAutomatic speech recognition 377
- mradermacher/indri-0.1-124m-tts-GGUF 300
- OpenMOSS-Team/MOSS-Transcribe-preview-2BAutomatic speech recognition 299
How to load it
from datasets import load_dataset
ds = load_dataset("speechcolab/gigaspeech")