Related Experiment Video
Updated: Jan 9, 2026
04:32
Sound Source Localization Testing in Single-sided Deafness Following Bone Conduction Intervention
Published on: December 20, 2024
786
Sound Source Localization for Autistic Children's Session Recordings
Summary
This study introduces a deep neural network for sound source localization in Autism Diagnostic Observation Schedule (ADOS-2) rooms. The method achieves 85-89% accuracy, aiding autism diagnosis by analyzing spatial behavior.
Area of Science:
- Acoustics
- Machine Learning
- Clinical Psychology
Background:
- Accurate sound source localization is crucial for analyzing spatial behavior in clinical settings like the Autism Diagnostic Observation Schedule (ADOS-2).
- Traditional beamforming methods are unsuitable for the non-uniform microphone arrays typically found in ADOS-2 rooms.
- Deep learning offers a promising alternative for sound source localization in complex acoustic environments.
Purpose of the Study:
- To develop and evaluate a novel deep neural network (DNN)-based approach for localizing simulated sound sources within an ADOS-2 observation room.
- To overcome the limitations of conventional methods imposed by unconventional microphone array configurations.
- To assess the potential of DNN-based sound source localization for enhancing autism diagnosis and treatment.
Main Methods:
- Formulated sound source localization as a classification problem, dividing the ADOS-2 room into four distinct zones.
- Developed two DNN architectures: a bidirectional long short-term memory (BiLSTM) network and a hybrid BiLSTM with a transformer encoder.
- Trained and tested models using simulated sound sources across diverse acoustic conditions and microphone array setups.
Main Results:
- Achieved classification accuracies ranging from 85% to 89% in various acoustic environments and with different microphone array configurations.
- Demonstrated the efficacy of the proposed DNN approach in achieving efficient sound source localization performance.
- Validated the potential of the method for practical application in clinical settings.
Conclusions:
- The proposed DNN-based sound source localization method is effective for ADOS-2 observation rooms with non-uniform microphone arrays.
- This technology can provide valuable insights into children's spatial behavior during autism evaluations.
- The approach holds significant potential for improving the accuracy of autism diagnosis and informing intervention strategies.
