Modeling Noise-Related Timbre Semantic Categories of Orchestral Instrument Sounds With Audio Features, Pitch...

Area of Science:

• Acoustic Musicology
• Psychoacoustics
• Machine Learning in Auditory Perception

Background:

• Audio features like inharmonicity and spectral roll-off correlate with 'noisy' sounds.
• These acoustic features likely influence multiple, semantically diverse timbre categories.
• Understanding timbre perception requires linking objective audio features to subjective listener experiences.

Purpose of the Study:

• To investigate the relationship between stimulus properties, audio features, and semantic timbre categories: raspy/grainy/rough, harsh/noisy, and airy/breathy.
• To explore how playing techniques and pitch register affect timbre perception, valence, and perceived exertion.
• To compare machine learning models for predicting semantic timbre ratings from audio features.

Main Methods:

• 153 participants rated 52 orchestral instrument sounds (varied families, registers, techniques) on three semantic categories, exertion, and valence.
• 44 summary audio features were extracted using the Timbre Toolbox (R-2021 A).
• Random Forest and Partial Least-Squares Regression models predicted semantic ratings from audio features.

Main Results:

• Extended techniques and pitch register significantly impacted valence, exertion, and raspy/harsh ratings; instrument family affected airy/breathy ratings.
• Random Forest models significantly outperformed Partial Least-Squares Regression in predicting semantic timbre ratings.
• Distinct audio feature patterns differentiate the raspy/grainy/rough, harsh/noisy, and airy/breathy semantic categories, despite feature overlap.

Conclusions:

• Audio features can predict semantic timbre perceptions, with non-linear models like Random Forest showing superior performance.
• Specific acoustic properties are uniquely associated with distinct semantic timbre categories, enabling their differentiation.
• The study advances understanding of timbre perception by linking objective acoustic measurements to subjective semantic interpretations.