Show HN: Voice-tracked teleprompter using on-device ASR in the browser

This project showcases the maturity of on-device inference in web browsers, combining local automatic speech recognition (ASR) with practical application design. The voice-tracked teleprompter uses client-side models to process speech entirely within the browser, eliminating privacy concerns and server dependencies while maintaining real-time responsiveness.

For local LLM practitioners, this is instructive on multiple fronts: it demonstrates that modern browsers can handle meaningful ML inference workloads, shows how to integrate ASR models into consumer applications, and provides a concrete example of privacy-preserving AI. The promptme-ai repository documents implementation details for developers interested in deploying ASR locally. Browser-based on-device inference is increasingly viable thanks to WebAssembly, ONNX Runtime, and frameworks like Transformers.js.

As users become more privacy-conscious and organizations seek to reduce dependency on external APIs, demonstrating that sophisticated tasks like speech recognition can run entirely locally is both a technical achievement and a validation of the local-first inference movement.

Source: Hacker News · Relevance: 8/10