This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
I built LinguaGrass, a translator that you use without looking at a screen. You wear a mic and an earphone, keep your laptop in your bag, and when someone speaks Hindi to you, you hear the English in your ear a few seconds later.
I made it because translation apps make you stare at your phone right when you should be looking at the person talking to you. This way you keep eye contact and actually enjoy where you are. It's for travelers, backpackers, and anyone who has been stuck in a market with no common language.
Code
akshvy
/
linguagrass
LinguaGrass is a screenless, eyes-free real-time translation system designed for travelers, backpackers, and cross-cultural explorers.
linguagrass
LinguaGrass is a screenless, eyes-free real-time translation system designed for travelers, backpackers, and cross-cultural explorers.
How I Built It
It's a simple pipeline in Python:
Mic → faster-whisper → Qwen 2.5 3B (Ollama) → ElevenLabs / Piper → Earphone
- faster-whisper listens and turns speech into text, locally.
- Qwen 2.5 3B (open-weight, running through Ollama) translates it, also locally.
- ElevenLabs speaks the translation when there's internet. If there's no signal, it switches to Piper, an open-source offline voice.
Right now it only does Hindi to English, and I tested it with everyday sentences like asking for directions or prices. It takes about few seconds per sentence on my laptop. [ONE REAL LIMITATION YOU NOTICED].
Why Does Open Innovation Matter?
Places where you need a translator most, like trails and village markets, often have no signal. Because Whisper and Qwen are open models, the main part of the project runs on my laptop with the internet off. A closed cloud API simply wouldn't work there.
It also keeps other people's voices on my device instead of sending them to someone's server, and it costs nothing to keep running.
Prize Categories
- Touch Grass (main theme)
- ElevenLabs
Top comments (0)