Local Speech Engines
Compare the local speech engines available in MachinesFluent and when to use them.
Local speech engines process speech recognition on your Windows computer. They are the right starting point when you want dictation without sending microphone audio to a cloud speech-to-text provider.
What Local Means
When a local speech engine is selected:
- microphone audio is handled by a model running on your computer
- speech recognition does not require a cloud STT provider
- the transcript can still be inserted into any app
- model files may need to download before first use
Local dictation does not automatically make every workflow local. If you send the resulting transcript to a cloud AI provider, that later AI request follows the provider path you selected.
Engine Guide
| Engine | Practical role | Hardware note |
|---|---|---|
| Parakeet | General dictation and everyday text entry. | Designed to work without requiring a dedicated GPU. |
| MedASR | Dictation where medical vocabulary support matters. | Designed to work without requiring a dedicated GPU. |
| Faster Whisper | Accuracy-focused transcription and multilingual workflows. | Best experience may depend on GPU acceleration and selected model size. |
| Kyutai Moshi | Low-latency or streaming-style transcription experiments. | Best experience may require a compatible NVIDIA GPU. |
Medical terminology is not medical compliance
MedASR support means the engine is useful for medical vocabulary. It does not mean MachinesFluent is certified for a regulated medical workflow. Review sensitive text before using or storing it.
Choosing An Engine
Choose based on the constraint that matters most:
- simplest private first run: start with a CPU-friendly local engine
- highest accuracy: test Faster Whisper against your own microphone and vocabulary
- medical terms: test MedASR with your real terminology
- low-latency experiments: test Moshi if your hardware supports it
Model Downloads
Some local models are downloaded on first use instead of being bundled in the installer. Keep MachinesFluent open during the first download and avoid switching networks mid-download.
After a model is available locally, the speech recognition path can run without a cloud STT provider. Other app features, such as updates, activation, cloud AI, cloud STT, or web search, may still use the network.