Dictating a message and recording a meeting may look similar: in both cases, speech becomes text. Their needs are different, though. A message needs a quick, usable sentence; a meeting may need speaker labels, an audio recording and a summary. FluidVoice offers dictation on macOS, while its FluidMeet section handles meetings. [1][2]
What happens in a meeting?
FluidMeet records locally, displays live captions and produces a transcript with speaker labels. You can edit text and export transcripts with timestamps and names. The release notes also describe summaries using a model on the Mac or a provider chosen by the user. [2]
In a The Next New Thing interview, the developer explains that the custom models shown at the time were for dictation and meetings used different models. That is a snapshot, not a permanent rule: the 2 November 1.6.13-beta.2 notes say compatible Fluid Speech models can also be selected for meeting transcripts. Beta availability and features depend on the build. [2][4]
Choose the model for your language
The developer describes Fluid Speech Mini as an English dictation model. Its Mac Core ML package is 253 MB; separate ONNX and GGUF model builds exist for other environments. Those model files do not make the FluidVoice application itself a Windows product: the project describes the app as available on macOS 15 or later, with Whisper support on Intel Macs. Check each model’s language support and version before choosing it. [1][3]
The Mini model card reports a 4.54% average word error rate across eight English test sets in the developer’s own runs. That is a useful comparison under those conditions, not a promise for your accent, meeting room or microphone. The developer notes stronger performance against background voices than against traffic noise. Published speed figures likewise describe specific tests and hardware. [3]
Where does the data go?
The project says voice and text stay on the machine when local processing is used. Selecting a cloud provider changes that boundary; giving an external agent access to meeting notes can also change where their contents travel. Optional screen context is off by default in the beta described by the release notes. Local processing alone cannot secure files you share or a device someone else can access. [1][2]
Try a short sample in your own language before relying on it. Check names, numbers and speaker labels, and inspect the summary provider. Tell meeting participants when you intend to record.
