⚠️🚧 This repository is under construction. Stay tuned! 🚧⚠️

News

• 16 June 2024 Added more details for EPIC-SOUNDS!
• 11 June 2024 Training scripts released!
• 10 June 2024 Setup guide released!
• 07 June 2024 Code released! (Initial release — further setup and cleaning in progress.)
• 06 June 2024 Checkpoints released!
• 05 June 2024 ArXiv Preprint released: https://arxiv.org/abs/2406.03344
• 22 April 2024 OpenReview Preprint released: https://openreview.net/forum?id=RZu0ZlQIUI

Overview

This repository contains the implementation of Audio-Mamba (AuM), a generic, self-attention-free and purely state space model designed for audio classification. It provides the necessary code for training and evaluating the model across various audio classification benchmarks. AuM is built on the works AST and ViM, and it utilizes Hugging Face's Accelerate library to facilitate efficient multi-GPU training.

Setting Up the Repository

Please run the following commands to set up the repository:

Create a Conda Environment

conda create -n aum python=3.10.13
conda activate aum

Setting Up CUDA and CuDNN

conda install nvidia/label/cuda-11.8.0::cuda-nvcc
conda install nvidia/label/cuda-11.8.0::cuda
conda install anaconda::cudnn

Installing PyTorch and Other Dependencies

pip install torch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

Enabling Bidirectional SSM Processing

To integrate the modifications for supporting bidirectional processing, copy the mamba_ssm folder to the site-packages directory of the Python installation within the Conda environment. This folder is directly borrowed from the ViM repository.

cp -rf vim-mamba_ssm/mamba_ssm $CONDA_PREFIX/lib/python3.10/site-packages

Inference

An example notebook for inference is provided in the examples/inference directory. The notebook demonstrates a minimal example of how to load a trained model and perform inference on a sample audio file.

Training

Overview

Each dataset's training scripts and relevant files are located within their respective folders under the exps/ directory. These folders include:

• Bash files: Represent configurations for various experiments conducted with the dataset.
• Data folder (except for EPIC-SOUNDS): May include example train/test splits, dataset label information, and data sampling weights.

Executing Training Scripts

To execute the training scripts:

1. Navigate to the dataset's directory (e.g., exps/vggsound/).
2. Run the corresponding script (e.g., bash aum-base_scratch-vggsound.sh).

Note: The scripts are prepared for execution but require modification of paths (such as experiment directories) to fit your specific setup.

Multiple GPU Training

For training on multiple GPUs:

1. Set GPU IDs: List the GPU IDs in the CUDA_VISIBLE_DEVICES environment variable (e.g., CUDA_VISIBLE_DEVICES=0,1,2,...).
2. Adjust Batch Size: Set the batch_size argument in the script to the desired batch size per GPU.

Note: To maintain the effective batch size from single GPU training, divide the batch size by the number of GPUs.

EPIC-SOUNDS Dataset

The EPIC-SOUNDS dataset has a distinct training structure:

• No data folder: The epic-sounds/ directory contains only training scripts.
• Config File: Additional setup details are available in the config_default.yaml file located at src/epic_sounds/epic_data/ directory. This file includes paths to the dataset folder, training splits, and other relevant default settings (please modify several of these variables according to your setup). Inside run.py file, some of the variables from this config file are overriden by the command line arguments.

For the full reference regarding this dataset, please refer to the EPIC-SOUNDS repository.

Model Checkpoints

The model checkpoints are available for the following experiments:

Base Scratch

These are the checkpoints for the base models with the variant Fo-Bi (b), trained from scratch.

Small ImageNet

These are the checkpoints for the small models with the variant Bi-Bi (c), initialized with ImageNet pretrained weights.

Base AudioSet

These are the checkpoints for the base models with the variant Fo-Bi (b), initialized with AudioSet pretrained weights.