Voiced
A dictation app for Linux, built by someone who actually uses it.
Show the engineeringHide the engineering
A voice writing assistant with pluggable ASR engines, evaluation-driven development, and AI text polishing. Built for Wayland with local-first processing.
Built with
- Python
- Whisper
- Wayland
- DBus
- ASR Engines
Live Preview
Experience Voiced
Case Study
The thinking behind Voiced
Problem
Linux lacks a reliable, high-quality dictation tool. Existing solutions are Windows/Mac-centric, tightly coupled to single ASR engines, and have no way to evaluate transcription quality systematically.
Solution
Created a pluggable architecture with multiple ASR engine adapters (Whisper, Deepgram, etc.), an evaluation-first development pipeline, and Wayland-native text insertion with fallback strategies.
Impact
Evaluation-first approach surfaced quality issues 3x faster than intuition-driven development. Supports multiple ASR backends through a common adapter interface.
Lilo