A Digital Twin-Based Speaker Placement Planning Tool for Indoor Environments

Nowadays, speakers are essential components for message delivery in indoor environments, including lectures, public addresses, and emergency announcements. Their physical placement should ensure adequate direct-sound audibility across occupant service areas while maintaining installation feasibility. A digital twin is a technology that allows an infrastructure layout to be designed and evaluated virtually on a computer before physical installation by reconstructing an indoor environment as a 3D model. In previous studies, we have proposed a method to reconstruct a 3D indoor model of an indoor environment from its 360∘ panoramic images using 3D Gaussian Splatting (3DGS) and a 3D point cloud, and applied it to surveillance camera placement. In this paper, we propose a digital twin-based speaker placement planning tool for indoor environments by generalizing the previous method to direct-sound acoustic simulation. This tool consists of four stages: (1) reconstructing a 3D indoor model from 360∘ panoramic images and extracting floor and desk receiver surfaces, (2) assigning the initial speaker budget based on the reconstructed floor area, (3) determining speaker mounting positions on valid ceiling regions using K-means spatial clustering under obstacle and boundary constraints, and (4) simulating broadband direct-sound sound pressure level (SPL) across floor and desk receiver surfaces. For evaluation, the proposed tool was deployed across three real-world indoor scenarios: a basketball hall (32.15 m×21.99 m), a furnished office (7.17 m×6.17 m), and a non-convex L-shaped office (23.00 m2). The experimental results showed that in each scenario, the generated layout achieved complete direct-sound audibility compliance across all sampled physical test locations (≥60 dB floor/≥65 dB desk) with zero detected hotspots exceeding 85 dB, maintaining SPL values between 66.91 dB and 77.88 dB. An on-site physical measurement campaign confirmed that the generated layouts satisfy target audibility thresholds under real room conditions, with mean absolute errors between 1.78 dB and 2.41 dB. These results confirm the practical utility of our approach as an initial geometry-driven planning tool for indoor audio infrastructure deployment.