Voice Cloning for Child Speaker Anonymization

This project explores the feasibility of using voice cloning as a solution for anonymizing the voices of child speakers. The aim is to protect the privacy of minors by altering their voices while retaining speech utility (including intelligibility and naturalness).

🛠️ Setup

Follow these steps to set up the environment:

1. Create a new conda environment:

   conda create -n child_speaker_anonymization python=3.10

2. Activate the environment:

   conda activate child_speaker_anonymization

3. Install required dependencies:

   pip install -r requirements.txt

If you use OpenMind, the data for our first experiments is stored here: <TODO: ADD>. Soon(ish) we will set up a datalad repo for taking care of data versioning (TODO).

🚀 Quick Start

To run the project, simply execute:

python main.py

📁 Project Files

Access the project files and related data here.

TODO:

-[] SETUP DATALAD WITH GOOGLE DRIVE (SUBDATASETS) - for now you can find it at /nese/mit/group/sig/projects/fabiocat/children_voice_anonymization

-[] VOICE CLONING PIPELINE

-[] ASR BENCHMARKING

-[] SER BENCHMARKING