SoniTranslate

You can run SoniTranslate using a Google Colab notebook, which provides a managed runtime environment. This is useful for users who do not want to set up a local environment.

https://colab.research.google.com/github/R3gm/SoniTranslate/blob/main/SoniTranslate_Colab.ipynb

To launch the SoniTranslate web interface:

1. Activate the environment:

   conda activate sonitr

2. Run the application script:

   python app_rvc.py

3. Open your browser and navigate to the local URL displayed in the terminal (typically http://127.0.0.1:7860).

Before installing SoniTranslate, ensure the following requirements are met:

1. NVIDIA Drivers: Install drivers for CUDA 11.8.0.
2. Pyannote License: Create a Hugging Face account and accept the license agreements for:
   • https://huggingface.co/pyannote/speaker-diarization
   • https://huggingface.co/pyannote/segmentation
3. Hugging Face Token: Create an Access Token on Hugging Face. Ensure you check "Read access to contents of all public gated repos you can access".
4. Anaconda/Miniconda: Install for Python environment management.
5. Git: Install Git (can be done via conda install -c anaconda git -y).
6. FFmpeg: Required for multimedia processing. Install via conda install -y ffmpeg and verify with ffmpeg -h.

You can extend SoniTranslate with additional Text-to-Speech capabilities:

Piper TTS (Fast, local neural TTS):

pip install -q piper-tts==1.2.0

Coqui XTTS (Realistic voice cloning):

pip install -q -r requirements_xtts.txt
pip install -q TTS==0.21.1  --no-deps

To stop the application, press Ctrl+C in the terminal. To deactivate the environment, run conda deactivate.

If you need to perform a clean reinstall, delete the SoniTranslate folder and remove the Conda environment:

conda deactivate
conda env remove -n sonitr

Follow these steps to set up the SoniTranslate environment (tested on Linux):

1. Create and activate Conda environment:

   conda create -n sonitr python=3.10 -y
   conda activate sonitr
   python -m pip install pip==23.1.2 Setuptools==80.6.0
   conda install pytorch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 pytorch-cuda=11.8 -c pytorch -c nvidia

2. Clone the repository:

   git clone https://github.com/r3gm/SoniTranslate.git
   cd SoniTranslate

3. Install dependencies:

   pip install -r requirements_base.txt -v
   pip install -r requirements_extra.txt -v
   pip install onnxruntime-gpu

You can extend SoniTranslate with additional Text-to-Speech capabilities:

Coqui XTTS (Voice cloning and realistic multilingual TTS):

pip install -q -r requirements_xtts.txt
pip install -q TTS==0.21.1  --no-deps

Piper TTS (Fast, local neural TTS): Note: Windows support is experimental.

pip install https://github.com/R3gm/piper-phonemize/releases/download/1.2.0/piper_phonemize-1.2.0-cp310-cp310-win_amd64.whl
pip install sherpa-onnx==1.9.12
pip install piper-tts==1.2.0 --no-deps

# Coqui XTTS
pip install -q -r requirements_xtts.txt
pip install -q TTS==0.21.1  --no-deps

# Piper TTS (Experimental on Windows)
pip install https://github.com/R3gm/piper-phonemize/releases/download/1.2.0/piper_phonemize-1.2.0-cp310-cp310-win_amd64.whl
pip install sherpa-onnx==1.9.12
pip install piper-tts==1.2.0 --no-deps

To launch the Gradio-based web interface:

1. Activate the environment:

   conda activate sonitr

2. Set your Hugging Face token as an environment variable:

   export YOUR_HF_TOKEN="YOUR_HUGGING_FACE_TOKEN"

3. Run the application:

   python app_rvc.py

4. Open the provided local URL (typically http://127.0.0.1:7860) in your browser.

Instead of exporting variables every session, you can set them permanently within the Conda environment.

Set Hugging Face Token:

conda activate sonitr
conda env config vars set YOUR_HF_TOKEN="YOUR_HUGGING_FACE_TOKEN_HERE"
conda deactivate

Set OpenAI API Key (for translation, TTS, or transcription):

conda activate sonitr
conda env config vars set OPENAI_API_KEY="your-api-key-here"
conda deactivate