Robust Audio Content Classification Using Hybrid-Based SMD and Entropy-Based VAD.

Area of Science:

• Signal Processing
• Machine Learning
• Acoustics

Background:

• Audio signals often contain a mixture of speech, music, and background noise.
• Variable noise levels pose a significant challenge for accurate audio content classification (ACC).

Purpose of the Study:

• To propose a robust hierarchical audio content classification (ACC) system.
• To enhance ACC performance in environments with variable noise levels.

Main Methods:

• A three-part hierarchical approach: entropy-based voice activity detection (VAD), hybrid feature extraction (1D-SEI and 2D-TII), and support vector machine (SVM) classification.
• Post-processing rules for segment refinement.

Main Results:

• The proposed system successfully classifies noisy audio into noise, speech, and music.
• Achieved comparable performance to existing methods across three datasets, even in variable noise conditions.
• Demonstrated improved audio content discrimination through the VAD scheme and hybrid features.

Conclusions:

• The hierarchical ACC system with entropy-based VAD and hybrid features offers a robust solution for audio classification.
• The proposed architecture effectively handles variable noise levels, enhancing classification accuracy.