TransVoix

The system captures audio from a phone call, converts it to text, translates the content into a target language, and then synthesizes the translation using a voice model that mimics the speaker’s own timbre. This pipeline operates continuously so that participants hear the translated speech in near real time, allowing a single conversation to flow across language barriers without switching to text‑based chat.

It is aimed at teams that interact with multilingual callers, such as healthcare providers, government agencies, and call‑center staff. By preserving the original speaker’s vocal characteristics, the tool helps maintain a personal tone while delivering accurate, on‑the‑fly translations during live calls.

The implementation is presented as an experimental prototype, with a live demonstration available on the project’s website. It focuses on end‑to‑end voice‑to‑voice translation rather than separate transcription or text translation services.