Pollyduble

Pollyduble

Automatic Dubbing with Voice Cloning and Speech Recognition
Made possible thanks to OpenVoice, MeloTTS, Faster Whisper, VoiceFixer, python-audio-separator and FFmpeg.

This is a highly experimental prototype of a script that aims to automatically dub English audio over a video file originally recorded in any language Whisper supports. Theoretically, with some modifications and different OpenVoice models, it should support any language supported by OpenVoice, however the translation would have to be handled by something other than Whisper

Features

• Voice cloning and local text-to-speech synthesis
• Automatic speech recognition
• Audio separation
• Automatic synchronization of dubbed lines to the original speech
• Optional voice fixing to bring back some high frequencies lost during the voice cloning process
• Muxing the dubbed audio and extracted instrumental track back into the video

---

PRs are welcome, this is mostly just a proof-of-concept. Some good ideas for improvement include:

• Speaker diarization to separate the speech of different characters and automatically assign the correct dubbed lines to the correct characters
• Ability to load custom subtitles instead of relying on automatic speech recognition
• A translation neural network (local is highly preferred) or API to not rely on Whisper's shoddy translations

Pre-requisites

• Python 3.9
• FFmpeg, FFprobe and FFplay installed on your system and in PATH
• Windows (only tested on Windows)
• A modern NVIDIA GPU with CUDA support is probably required
• Miniconda or Anaconda (optional, but recommended)

Installation

0. Install FFmpeg, FFprobe and FFplay on your system and make sure they are in PATH. You can download them from here.

1. Make a new directory and clone this repository:

git clone https://github.com/igerman00/Pollyduble
cd Pollyduble

2. Create a new Conda environment:

conda create -n dubbing python=3.9

3. Activate the Conda environment:

conda activate dubbing

4. Clone the OpenVoice repository

git clone https://github.com/myshell-ai/OpenVoice

Make sure the OpenVoice repository is in the same directory as this repository, it should be named "OpenVoice".

5. Install OpenVoice:

cd OpenVoice
pip install -e .
pip install git+https://github.com/myshell-ai/MeloTTS.git
python -m unidic download

6. Install torch with GPU support (the index-url parameter should be optional for no GPU support):

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

7. Install the other dependencies:

cd .. # Go back to the root directory of the repo
pip install -r requirements-win-cu118.txt

Usage

1. Obtain a video file and place it anywhere on your computer, for this example we will assume it is in the same directory as our demo.py script, and it is named video.mp4.
2. Run the script:

python demo.py -i video.mp4 -s -m

The output will be stored in the Pollyduble/output directory by default. It will contain various files including the dubbed video, the separated audio, the dubbed audio, and the voice sample. Mostly, it should be one-click.

Options include:

• -i or --input to specify the input video file
• -o or --output to specify the output directory (default is Pollyduble/output)
• -v or --voice to specify a custom sample for the voice cloning. If not specified, one will be created from the first 15 seconds of the video
• -s or --separate to enable audio separation, i.e. extracting the background music and the speech from the video separately
• -m or --mux to enable muxing the separated audio back into the video with the dubbed speech
• -f or --fix to enable voice fixing, i.e. improving the quality of the dubbed speech.
  ^ Experimental and doesn't actually sound that good most of the time.
• --help to display the help message

License

This project is licensed under the MIT License - see the LICENSE file for details.