ALTIC develops macOS software including FluidVoice, a dictation application with local speech recognition and an optional local enhancement model. The creator also works on smaller models for polishing dictated text.[1][3]
Start with FluidVoice →
Contributions
Separate recognition from rewriting
FluidVoice converts speech into text with a choice of speech models, then can apply an optional enhancement step for formatting and cleanup. It inserts the result into the active application through macOS accessibility features. Local recognition and the Fluid Intelligence runtime keep that processing on the machine; users can instead enable a cloud provider for enhancement.[2]
An open app with an optional private runtime
The application is GPLv3 software, while Fluid Intelligence is a separately maintained private runtime. The repository distinguishes those components and documents required microphone and accessibility permissions. Its privacy notes also distinguish voice content from anonymous activity signals and optional detailed analytics.[2]
Ideas
In a discussion of memory use, ALTIC describes experimenting with smaller enhancement models and different hosting arrangements. A user’s preferred model, available memory and tolerance for loading delays affect how useful dictation feels in practice. The creator discusses adding a toggle to release the model after inactivity.[3]
Tools, writing & talks
Tools
Start here
FluidVoice
github.comA macOS dictation app with selectable speech engines, text insertion and optional local or cloud enhancement.
Read the speech-model table and privacy notes before choosing a configuration. Trying dictation needs a supported Mac plus microphone and accessibility permissions; model and enhancement choices affect memory, downloads and cloud use.
Creator and maintainer[1][2]Original source