A Survey of Optimization Methods for Independent Vector Analysis in Audio Source Separation.

Area of Science:

• Artificial Intelligence
• Signal Processing
• Big Data Analytics

Background:

• Blind Source Separation (BSS) is crucial for extracting signals from mixed data.
• Independent Component Analysis (ICA) is a common BSS method, but has limitations.
• Independent Vector Analysis (IVA) extends ICA for multi-channel signals, resolving ambiguities.

Purpose of the Study:

• To provide a comprehensive survey of Independent Vector Analysis (IVA).
• To focus on optimization techniques for IVA update rules.
• To evaluate the performance of different IVA methods.

Main Methods:

• Review of basic principles of BSS, ICA, and IVA.
• Focus on existing IVA-based optimization update rule techniques.
• Experimental evaluation of IVA methods in deterministic and overdetermined environments.

Main Results:

• AuxIVA-IPA demonstrated the best performance in deterministic environments.
• AuxIVA-IP2 also showed strong performance in deterministic settings.
• OverIVA-IP2 achieved the best results in overdetermined environments.
• IVA-NG method's performance was suboptimal across tested environments.

Conclusions:

• IVA is a powerful extension of ICA for complex signal separation.
• Optimization of IVA update rules is key to enhancing performance.
• Specific IVA variants excel in different environmental conditions, guiding method selection.