Cloud dictation proved steadier than running speech recognition on a Mac
A local model was tested as an alternative to a dictation service costing about $15 per month. Even on a MacBook Pro with an M4 chip and 48GB of memory, the had glitches, was unreliable, and could not be used on a phone. A small API was then built with FastAPI so Apple Shortcuts could send recordings to OpenAI's Whisper service.
This approach worked much more reliably. Because the first were poorly formatted, was added to clean up and organize the text. The finished setup provided formatted dictation on a Mac, iPad, and iPhone.
On the iPhone, the shortcut was linked to the Action button for quick access.
Key points
- Local was unreliable on an M4 MacBook Pro with 48GB of memory.
- FastAPI connected Apple Shortcuts to OpenAI's Whisper service.
- cleaned up poorly formatted .
- The resulting dictation tool worked across Mac, iPad, and iPhone.
- The iPhone Action button launched the dictation shortcut directly.