Dataset · ai4bharat
Multilingual TTS for Indian Languages
ai4bharat/Rasa
Audio and text for expressive speech synthesis in six Indian languages.
What's inside
Rasa is the first high-quality multilingual expressive TTS dataset for Indian languages. The card specifies a minimum of 20 hours per speaker and aims to cover both female and male voices. The project is supported by Bhashini, EkStep Foundation, and Nilekani Philanthropies; languages include Assamese, Bengali, Kannada, Malayalam, Marathi, and Nepali.
What you can do with it
- Train a multilingual TTS model
- Fine-tune a voice for a specific speaker
- Evaluate synthesis expressiveness in Indian languages
- Tasks
- Text → speech
- Languages
- Assamese, Bangla, Kannada, Malayalam, Marathi, Nepali, Tamil, Punjabi and 4 more
- Size
- 1M–10M rows
- Downloads this month
- 13.3K
- Likes
- 73
- Access
- on request on Hugging Face
Models trained on it
- ai4bharat/IndicF5Text → speech 42K
- SPRINGLab/Indic-MioText → speech 4K
- Aditya02/IndicF5Text → speech 3.1K
- rsolanki1822/IndicF5-mirrorText → speech 1.3K
- tinisoft/piper-ta_IN-rasa_female-mediumText → speech 1K
- sulabhkatiyar/indic-tiny-tts-hindiText → speech 885
- vertexvicky/piper-rasa-tamilText → speech 734
- Phoonnx/f5tts-multi-spring-unicodeText → speech 594
How to load it
from datasets import load_dataset
ds = load_dataset("ai4bharat/Rasa")