Toward the Definition of a Soundscape Ranking Index (SRI) in an Urban Park Using Machine Learning Techniques.

Area of Science:

• Ecology
• Bioacoustics
• Machine Learning

Background:

• Assessing habitat acoustic quality is crucial for ecological monitoring.
• Existing methods may not fully capture the complex interplay of sound sources.
• A quantitative index is needed for effective environmental soundscape evaluation.

Purpose of the Study:

• To develop and validate a Soundscape Ranking Index (SRI) for accurate acoustic quality assessment.
• To optimize SRI weights using machine learning algorithms.
• To evaluate the contribution of biophony and anthropophony to the overall soundscape.

Main Methods:

• Trained four machine learning algorithms (Decision Tree, Random Forest, AdaBoost, SVM) on labeled sound recordings.
• Extracted spectral features including ecoacoustic indices and Mel-frequency cepstral coefficients (MFCCs).
• Focused labeling on identifying biophony and anthropophony within sound recordings from 16 sites in Parco Nord, Milan.

Main Results:

• Decision Tree and AdaBoost models achieved good classification performance (F1-score = 0.70, 0.71).
• The models successfully assigned differential weights to natural (biophony) and anthropogenic sounds.
• Optimized weights provided a reliable estimation of mean SRI values across sites.

Conclusions:

• Machine learning effectively optimizes weights for the Soundscape Ranking Index (SRI).
• The SRI serves as a valuable ecological tool for acoustic quality assessment in complex habitats.
• This approach supports both rapid on-site and remote ecological surveys.