Iceberg: Hybrid Auralization for Minimalist Setups

Iceberg is a hybrid spatial audio engine designed to deliver clinical-grade virtualization using minimal hardware. By merging the localization precision of VBAP with the immersion of Ambisonics, Iceberg allows researchers to create realistic, "ecological" acoustic environments with as few as four loudspeakers.

---

The Problem

Individuals with hearing loss struggle in noisy, social environments. Testing their performance in these scenarios usually demands a complex laboratory. Standard virtualization methods like Ambisonics or VBAP have trade-offs. Ambisonics offers great immersion but lacks sharp localization at low orders. VBAP provides precise directionality but often lacks the "feel" or spaciousness of a real room.

The Solution

Iceberg combines the strengths of both. It splits the Room Impulse Response (RIR) into two distinct parts:

1. Localization (Direct Sounde and Early Part): Handled by VBAP. This ensures the listener can pinpoint exactly where a sound starts.
2. Immersion (Late Part): Handled by Ambisonics. This provides the lush, reverberant "feel" of a real room.

The transition point isn't arbitrary. Iceberg uses Center Time ($T_s$) to find the ideal moment to switch from one method to the other.

---

Key Features

• Minimal Hardware: Optimized for small reproduction systems (4+ speakers).
• Dual-Listener Support: Validated for up to two participants without losing spatial accuracy.
• Clinically Validated: Proven effective in speech-in-noise tasks involving both normal-hearing and hearing-aided participants.
• High Accuracy: Maintains localization cues within a 30° ambiguity angle in the horizontal plane.

How it Works

The system uses MATLAB to process signals through a calibrated setup. By using $T_s$ to divide the impulse response, the early reflections that dictate spatial origin are panned precisely. The remaining energy is encoded via first-order Ambisonics to fill the room.

---

Quick Start (For Developers)

1. Prerequisites

This project requires MATLAB and several specialized toolboxes. Ensure the following are in your path:

• ITA Toolbox
• Higher-Order-Ambisonics
• Spherical-Harmonic-Transform
• Vector-Base-Amplitude-Panning
• SAP Voicebox

2. Implementation

• Calibration: Navigate to /Calibration and run calibration_LSRoom.m. Update this with your specific loudspeaker coordinates.
• Generate Audio: Run iceberg_example.m to see how to take a dry signal and an RIR to create a spatialized output.

---

Thesis & Research

This method was developed as part of a deep dive into listening effort and virtualization. Download the full Thesis (PDF)

Key Chapters

• Chapter 3: Binaural Cue Distortions – A comparison of VBAP and Ambisonics through a calibrated setup, examining spatial distortions and the impact on a second listener.
• Chapter 4: Behavioral Study – An investigation into how signal-to-noise ratio (SNR) and reverberation impact listening effort, using EEG and subjective questionnaires.
• Chapter 5: The Iceberg Method – The formal proposal and evaluation of the hybrid method using objective parameters and hearing aid verification.

Core Findings

• Center Time ($T_s$): Successfully identifies the transition point between early and late reflections to split impulse responses.
• Immersion vs. Accuracy: The hybrid method matches the localization accuracy of VBAP while maintaining the sense of immersion typically only found in Ambisonics.
• Clinical Viability: The setup provides reliable binaural cues within a 30° ambiguity angle, making it suitable for audiological tests in smaller clinic rooms.