Inspiration
I ran into https://www.augmental.tech/vox which I thought was really cool! But not cool for the $200 price tag
What it does
Vox internal microphone for voice controls using humming, singing, or clicks with swipe gestures and pitch controled cursor
How we built it
- Pico Pi 2w with bluetooth in 3d printed case
- software and mobile app for pitch variations and tones (With calibration options)
- Custom fine-tuned System One classifer model (Verdict) ## Challenges we ran into I had to figure out how to account for the variety of human voices. So I set up an agentic loop to automatically train on an emulated android phone clicking through apps and swiping. I originally intended for the System One model to be the thing classifying swipes and such, but I realized it was much simpler having deterministic rules. ## Accomplishments that we're proud of I built a cute little character and like the end design. Training and fine tuning an AI model was not what I planned, but going through and doing so it actually was more return on the time invested in training then I expectes. ## What we learned
- I learned how to fine tune an AI model
- Had to figure out mobile app development with help of AI
- Managing variety of datasets
What's next for Canti
I will likely be continuing to fine tune ans improve the latency of Canti and open source it
Log in or sign up for Devpost to join the conversation.