ASD-Detection

This repository provides the main implementation used in our manuscript for audio-based ASD classification, including the proposed deep learning model, conventional machine learning baselines, and SHAP-based interpretability analysis.

Repository Structure

• AudioTransformer.py
  Implementation of the main Transformer-based audio classification model used in this study.

• train_audio_transformer.py
  Training and evaluation script for the Transformer-based deep learning experiments.

• train_machine_learning_model.py
  Script for the conventional machine learning baseline experiments.

• ShapAnalysis.py
  Script for SHAP-based feature importance analysis on handcrafted acoustic features.

• dataset.py
  Dataset construction and loading utilities used by the above experiments.

Data Availability

Due to privacy and ethical restrictions associated with the human-subject audio dataset, the raw training and test audio files are not publicly released at this time.

Minimal Example

A minimal forward-pass example is provided in AudioTransformer.py to illustrate the expected input format and model usage:

import torch
from AudioTransformer import AudioClassifier

x = torch.randn(1, 1, 16000)
model = AudioClassifier()
y = model(x)
print(y.shape)