"Zero-Touch Music Detection and Classification for Internet of Audio Things under Real-World Acoustic Conditions"

TISCode is a system that transmits short digital messages through AI-generated music clips over the acoustic channel, requiring no wireless infrastructure. The core challenge is the music classification gate — the component that decides whether captured audio is a valid transmission before decoding. This thesis evaluates three gate models — YAMNet, VGGish, and a lightweight CNN under clean audio, synthetic noise, and real-world noise conditions, and validates the system through field tests in indoor, outdoor, urban, and healthcare environments. YAMNet proved the best choice with 98.8% accuracy and only 1% false rejection rate. A custom lightweight CNN with just 8,000 parameters, when trained with data augmentation, approached YAMNet's performance while being 400 times smaller making it viable for edge IoT devices.