The Nipun Lakshya app, used in Uttar Pradesh for assessing students’ oral fluency via Bhashini APIs, faces challenges due to unreliable internet, leading to failed assessments and incorrect scores. To enable offline functionality, we reduced the model size from 460 MB to 140 MB using quantization (both static and dynamic). We further optimized the model with PyTorch for mobile compatibility and CPU performance. This will now allow for Bhashini’s ASR model to work offline with the Nipun Lakshya App.
About Contributor
" Aryan is pursuing a B.Tech in Information Technology & Mathematical Innovations at Delhi University. He is dedicated to making meaningful contributions to DPGs and has a passion for capturing the beauty of nature through photography. His enthusiasm for blending technology with creativity drives his academic and personal projects."
Key Impact Takeaways:
- This project’s output will become a new offering for Bhashini.
- Contributor Aryan is continuing on as a long-term intern with the Bhashini.