How local call translation works.

1. Choose the audio you need

Select a Windows audio device or supported application capture source. Play speech through that source and check that captions correspond to the intended speaker. Per-app capture is a source-selection feature; it does not install a translation bot into Zoom, Teams or another meeting service.

2. Recognize speech on your PC

The recognizer turns incoming audio into original-language captions. The current release includes Parakeet on the CPU; compatible NVIDIA systems can add Whisper large-v3. The app shows a model only when its required files and runtime are available. Compare the two recognition choices.

3. Translate captions locally

A local translation model turns the recognized text into the selected target language. Original and translated captions remain visible for comparison. The application displays text; it does not synthesize translated speech, replace anyone’s voice or send a translated audio track to the other caller.

What offline means

Recognition and translation run locally once compatible packs are installed. Initial setup, downloads and optional licensing, purchase or support services can require internet access. Local processing avoids per-minute cloud translation charges, but your PC still needs enough memory and processing capacity.

Accuracy and timing

Background noise, accents, overlapping speakers and names or numbers can cause errors. Drafts may change before captions finalize, and finalized captions can still be wrong. Translation cannot reliably recover a number that recognition omitted. Confirm important details with the speaker.