When local transcription makes sense
The built-in SenseVoice option runs speech recognition on your computer after its model has been downloaded. That means the speech-to-text step can work without an internet connection. It is a useful choice when your connection is unreliable or you want audio to stay on your device for transcription.
Local processing still uses your machine's resources, and the model must be downloaded first. Check the output yourself when accuracy is important.
When a cloud engine makes sense
Cloud providers offer another set of speech models and may fit your language, accuracy, or speed needs. RediFinch supports several providers through its Engine menu and stores the API key you supply in local app settings.
Cloud transcription sends audio to the selected provider and depends on your connection. Check that provider's terms, retention policy, and pricing before using it for sensitive work.
Keep the whole pipeline in mind
A local speech engine does not make every output mode offline. If you enable AI polishing, AI translation, a translation service, or the voice agent, the relevant text or tool results go to the provider you configure.
For an offline transcription workflow, use the local engine with raw output. For a polished or translated result, decide separately which service should process the text.
A quick decision guide
Start with local transcription if offline use and control of the audio path matter most. Try a cloud engine if you want to compare model quality for your voice and language. Keep one short test phrase and one realistic paragraph so you can judge each setup on your actual work rather than a generic benchmark.
- Choose Local or a cloud provider in Settings → Engine.
- Dictate the same sample in both setups.
- Review punctuation, names, latency, and the provider's data policy before settling on one.