Dev building fully on-device iOS voice notes finds Apple models fine at STT, weak at summarization
IanAndrewsDC · x · 2026-10-19
A developer is building an iOS voice transcription and note-taking app that runs entirely on-device. Apple's built-in models handle speech-to-text at reasonable quality, but the same models produce poor output for summarization and action-item capture despite prompt tuning. They're asking whether others are working on similar setups.
Takeaway: on-device Apple models are now good enough for transcription, but generative tasks like summarization still lag cloud models noticeably, limiting fully offline note-taking products.
More from Apps
- Taliri adds learner-friendly pronunciation respellings for Swedish, German, Hindi — theteknosaur · 2026-10-19
- Luke Wroblewski charts 18 years of iOS calculator design evolution — LukeW · 2026-10-19
- Real-time portal jumping through Gaussian Splat worlds with WebGPU — XRarchitect · 2026-10-19
- Perplexity Computer builds an interactive LOTR Middle-earth map from scratch — AravSrinivas · 2026-10-19
- Tolan raises $30M from USV and Khosla for its AI companion app with 250K users — soleio · 2026-10-19
- Dev uses Claude Opus to build and open-source Orchard, a fully local semantic photo search app — chaseleantj · 2026-10-19