Our Open Source Contributions
Building foundational AI infrastructure for India, openly available for everyone.
🤗Hugging Face250k+ Downloads
svara-TTS
Open, expressive multilingual TTS for the next billion voices.
A speech foundation model speaking 19 Indian languages with natural rhythm and emotion. Built like a language model for speech — easy to fine-tune with just a few hours of audio.
19Languages
2000+Hours of Speech
~50Voice Speakers
Emotion Conditioning
😊Happy
😢Sad
😠Anger
😨Fear
😲Surprise
🔊Clear
19 Languages in Native Scripts
हिन्दीবাংলাमरাठीతెలుగుಕನ್ನಡதமிழ்മലയാളംગુજરાતીਪੰਜਾਬੀ+10 more
GitHub
Indic Text Normalization
Deterministic, low-latency normalization for 19 Indian languages.
A comprehensive WFST-based library built on Pynini that converts numbers, dates, currency, measurements and more into natural spoken form. Designed for TTS, ASR, and NLP pipelines. An extension of NVIDIA NeMo for Indic languages.
Normalization latency
5ms
WFST deterministic traversal
vs. LLM 500ms+100x
12 Semiotic Classes — examples
Cardinal Numbers25 → पच्चीसCurrency₹500 → पांच सौ रुपयेDates15/08/2024 → पंद्रह अगस्तTime10:30 → साढ़े दस बजेMeasurements5kg → पांच किलोग्रामFractions½ → आधा+6 more classes