← Back to projects
Audio · local AI
Year 2026
Magic Whisper
Exploring local speech-to-text that respects the machine and the person using it.
Overview
The work combines audio capture, voice activity detection, local ASR, post-processing, and safe text insertion. The goal is a dictation tool that feels immediate without sending the recording away.
Focus
Local audio and AI
Built with
SwiftCore MLFluidAudioSpeechAnalyzer
What I worked on
- On-device transcription and model loading
- Voice activity and recording health
- Context, cleanup, and safe insertion