An Empathy Evaluation System Using Spectrogram Image Features of Audio

Area of Science:

• * Music and Emotion Analysis
• * Affective Computing
• * Advertising Technology

Background:

• * Online video consumption is prevalent, with music significantly influencing viewer emotions and perceptions.
• * Understanding the link between music's physical properties and empathy is crucial for effective advertising.
• * Previous research has not fully explored specific audio features that elicit empathy.

Purpose of the Study:

• * To analyze music features in advertising videos that induce empathy.
• * To identify specific audio characteristics correlating with emotional resonance.
• * To develop a real-time advertisement empathy monitoring system.

Main Methods:

• * Combined Mel-frequency cepstral coefficients (MFCC) power spectrum analysis with image RGB analysis.
• * Utilized spectral analysis to categorize eigenvectors from low (blue) to high (red) frequencies.
• * Employed a random forest classifier for data classification and model training.

Main Results:

• * Achieved high accuracy with a trained model: 99.173% training accuracy and 86.171% test accuracy.
• * Identified low-frequency, high-amplitude audio as more empathetic than high-frequency, high-amplitude audio.
• * Determined frequency and audio amplitude as critical waveform attributes.

Conclusions:

• * Spectrogram image features of audio offer the most potent method for inducing human empathy.
• * The developed method provides a novel approach to audio extraction for empathy induction.
• * Findings are valuable for advertisers seeking to enhance emotional connection with audiences.