Modeling the onset advantage in musical instrument recognition

Area of Science:

• Acoustics
• Psychoacoustics
• Machine Learning

Background:

• Sound onsets are crucial for human musical instrument identification.
• The reason for this "onset advantage" remains debated: enhanced perception or richer acoustic data.

Purpose of the Study:

• To investigate whether the acoustic richness of sound onsets or enhanced perceptual encoding drives the onset advantage in musical instrument identification.
• To model human performance in a previous study on instrument identification from tone excerpts.

Main Methods:

• A Hidden Markov Model (HMM) classifier was employed.
• Separable Gabor filterbank features were used to process audio signals.
• The model simulated human performance in identifying musical instruments from tone excerpts.

Main Results:

• The HMM classifier successfully replicated the "onset advantage" observed in human listeners.
• The model's performance suggests that distinct acoustic qualities of onsets are key.
• This supports the hypothesis that acoustic richness, rather than solely perceptual encoding, drives the effect.

Conclusions:

• The "onset advantage" in musical instrument identification is likely driven by the distinct acoustic information present at sound onsets.
• Computational modeling can effectively simulate and explain psychoacoustic phenomena.
• Further research can explore specific acoustic features contributing to this advantage.