Skip to content
All work
AI & LLM2025

dubAI — Voice-Cloning Dubbing Pipeline

A FastAPI service that dubs video into another language while keeping the original speaker's voice.

PythonFastAPICoqui XTTS v2WhisperFFmpeg
Source

The problem

Standard dubbing replaces the speaker with a stranger. The information survives; the person doesn't.

How it works

The pipeline chains transcription, translation, and synthesis behind one FastAPI service: audio is extracted and transcribed, the text is translated, then Coqui XTTS v2 re-synthesises it in a voice cloned from the original speaker.

Each stage is a separate step in the pipeline, so a model can be swapped at any point without rewriting the service around it.

Outcomes

  • End-to-end dubbing from a single API call.
  • Speaker identity preserved across languages through voice cloning.