A new open-source speech recognition system called Swanada has been developed to address the challenges of on-device ASR across multiple Indic languages and Classical Sanskrit. The system significantly reduces storage requirements by using shared encoder models and script-specific linear CTC heads, cutting down the total size from over 2.1 GB to under 184 MB. Swanada achieves competitive accuracy with low latency and real-time factor, making it suitable for browser, Node.js, and Python environments. AI
IMPACT Enables efficient on-device speech recognition for a wide range of Indic languages, reducing costs and latency for applications.
RANK_REASON The item describes a new open-source system and its technical details, fitting the research category. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- Classical Sanskrit
- Conformer
- Indian English
- Indic Languages
- Node.js
- Python
- Sanskrit
- Swanada
- WebAssembly
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →