Robust automatic train pass-by detection combining deep learning and sound level analysis.

Area of Science:

• Acoustics and Signal Processing
• Environmental Noise Monitoring

Background:

• Growing demand for controlling high noise levels necessitates advanced automatic sound event detection.
• Limited research exists on automatic train pass-by detection, despite its significant annoyance factor.

Purpose of the Study:

• To develop an innovative and accurate method for automatic train pass-by detection.
• To improve the estimation of railway noise contribution in diverse soundscapes.

Main Methods:

• A generic classifier identifies vehicle noise from raw audio signals.
• Mel-spectrogram analysis and sound level metrics refine detection to isolate train pass-bys.
• The method was tested on various long-term audio signals.

Main Results:

• Achieved a 90% temporal overlap with reference demarcations for train pass-by events.
• Demonstrated high detection rates on diverse, long-term audio recordings.
• The developed technique effectively distinguishes train noise from other vehicle sounds.

Conclusions:

• The proposed method offers a reliable solution for automatic train pass-by detection.
• High detection accuracy facilitates precise railway noise contribution assessment.
• This advancement supports better management of noise pollution in urban and rural soundscapes.