dGSLM

dGSLM is a project for textless speech modeling using Fairseq. This guide will help you set up the environment and install the necessary dependencies.

---

Environment Setup

Follow the steps below to set up the environment using conda and pip:

1. Create and activate a Conda environment

conda create -n dGSLM python=3.9
conda activate dGSLM

2. Install PyTorch and dependencies

Install PyTorch, torchvision, and torchaudio with CUDA 12.1 support:conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia

3. Upgrade pip and install required packages

pip install pip==23.3.1
pip install omegaconf==2.0.6

4. Install Fairseq

Clone the Fairseq repository and install it in editable mode:

git clone [email protected]:facebookresearch/fairseq.git
cd fairseq
pip install --editable ./

5. Install additional dependencies

pip install soundfile librosa

6. Run Inference

python src/inference.py