Implementación de un sistema embebido para la clasificación de fuentes de ruido ambiental urbano utilizando técnicas de TinyML en un entorno acústico de Bogotá
Thesis
The increase in urban noise, stemming from the continuous development of economic and social activities in cities, has become a daily concern with a negative impact on the population. Studies presented by the District Secretary of Environment (SDA) of Bogotá, Colombia, indicate that at least 11.8% of the population is exposed to noise levels that exceed the recommendations of the World Health Organization. The main objective of this project is to design an intelligent classification tool that allows the identification and categorization of different noise sources in real-time. The implemented system is based on low-power devices equipped with audio sensors and processing capability (TinyML). The YAMNet model was used, optimized for the specific conditions of Bogotá, achieving accurate classification of noise sources into categories such as alarms, ambiance, applause, airplanes, human activities, impacts, motorcycles, and heavy vehicles. The results obtained show that in the acoustic environment under study, noises from heavy vehicles and motorcycles constitute a large part of the environmental noise in the sector. Additionally, airplanes, although less frequent, show that many events (87%) exceed the maximum permissible standard for the sector, reaching levels of up to 88.4 dBA. In conclusion, this work demonstrates that the use of TinyML for the classification of urban noise sources is a viable and effective strategy. The developed methodology facilitates more efficient management of urban noise, providing a solid foundation for future research and technological developments, with the potential to significantly improve the quality of life in urban environments.
