Dataset · ylacombe
Expresso Expressive English Speech
ylacombe/expresso
40 hours of expressive English speech: 11 hours of reading in 8 styles and improvised dialogues in 26 styles.
What's inside
Expresso is a high-quality (48 kHz) audio dataset in English. It includes expressive reading in 8 styles (mono WAV) and improvised dialogues in 26 styles (stereo WAV); a total of 4 speakers, 2 males and 2 females, around 40 hours. Transcripts are available for the read speech.
What you can do with it
- Training an expressive TTS model
- Evaluating ASR on expressive speech
- Analyzing style and emotion in speech
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 5.4K
- Likes
- 102
Models trained on it
- SPRINGLab/Indic-MioText → speech 4K
- 11mlabs/indri-0.1-124m-ttsText → speech 982
- parler-tts/parler-tts-mini-expressoText → speech 723
- mradermacher/indri-0.1-124m-tts-GGUF 300
- mradermacher/indri-0.1-350m-tts-GGUF 262
- MIT-SLS/USAD2-LargeEmbeddings 198
- shangeth/Wren-TTS-0.5B-multi-expressiveText → speech 155
- MIT-SLS/USAD2-SmallEmbeddings 71
How to load it
from datasets import load_dataset
ds = load_dataset("ylacombe/expresso")