For April Fools’, I built a fun demo called QuackToSQL — you quack into your mic, and it transcribes your “quack” into a SQL query, entirely in the browser.
It uses transformers.js (Hugging Face) with WebGPU for real-time transcription using Whisper, all running locally with no server-side code. I wrote up a deep dive into how it works, what WebGPU enables, and the future of local browser-based AI.
Also touches on challenges like model size, browser compatibility, and why this shift might matter beyond just fun demos.
It uses transformers.js (Hugging Face) with WebGPU for real-time transcription using Whisper, all running locally with no server-side code. I wrote up a deep dive into how it works, what WebGPU enables, and the future of local browser-based AI.
Also touches on challenges like model size, browser compatibility, and why this shift might matter beyond just fun demos.