On the Performance of Sound Bubbles: An Analysis of Directional Performance and Underlying Acoustic Principles: An Analysis of Directional Performance and Underlying Acoustic Principles

Abstract

Dette speciale har til formål at undersøge ydeevnen af det lydsystem, der er foreslået i artiklen "Hearable Devices with Sound Bubbles". Analysen fokuserer på tre centrale mål: (1) hvordan skarpheden af en lydboble afhænger af den konfigurerede boblestørrelse, (2) hvordan systemets ydeevne varierer med vinklen mellem lytteren og taleren, og (3) om det neurale netværk, der anvendes i artiklen, implicit udnytter den inverse afstandslov til at estimere afstanden til en taler. Arbejdet omfatter en teknisk gennemgang af den originale artikel, en MATLAB-baseret simulering baseret på den inverse afstandslov til at modellere idealiseret bobleadfærd, samt en systematisk evaluering af det trænede neurale netværk ved hjælp af specialudviklet testdata. Denne testdata blev genereret under forsimplede forhold for at isolere afstand og vinkel som variable, under antagelse af næsten anekoiske omgivelser og med enkelt-taler testscenarier. Resultaterne indikerer, at den inverse afstandslov spiller en rolle i det neurale netværks bagvedliggende mekanismer, idet præstationen i høj grad ligner resultaterne fra MATLAB-simuleringen. Dog skal disse resultater tolkes med forsigtighed, da testopsætningen er forsimplet – med lav variation i omgivelserne, ingen efterklang og idealiserede signalantagelser. Mere realistiske testforhold er nødvendige for at kunne vurdere, om systemet kan generaliseres til brug i virkelige omgivelser.

This master thesis investigates the performance of the audio system proposed in the paper Hearable Devices with Sound Bubbles. The analysis focuses on three key objectives: (1) how the sharpness of the sound bubble boundary depends on the configured bubble size, (2) how system performance varies with the angle between the listener and the speaker, and (3) whether the neural network employed in the paper implicitly leverages the inverse distance law to estimate speaker distance. The work includes a technical review of the original paper, a MATLAB-based simulation using the inverse distance law to model idealized bubble behaviour, and a systematic evaluation of the trained neural network using custom-generated test data. The custom data was created under simplified conditions to isolate distance and angle as variables, assuming near-anechoic environments and using single-speaker test cases. The results suggest that the neural network does utilize cues consistent with the inverse distance law, as its performance patterns closely match those of the MATLAB simulation. However, due to simplifications in the test setupincluding low environmental variation, no reverberation, and idealized signal assumptions, the findings must be interpreted with caution. More realistic testing conditions are needed to assess the generalizability of the system to realworld deployments.

Colophon: This page is part of the AAU Student Projects portal, which is run by Aalborg University. Here, you can find and download publicly available bachelor's theses and master's projects from across the university dating from 2008 onwards. Student projects from before 2008 are available in printed form at Aalborg University Library.

If you have any questions about AAU Student Projects or the research registration, dissemination and analysis at Aalborg University, please feel free to contact the VBN team. You can also find more information in the AAU Student Projects FAQs.

A master thesis from Aalborg University

On the Performance of Sound Bubbles: An Analysis of Directional Performance and Underlying Acoustic Principles: An Analysis of Directional Performance and Underlying Acoustic Principles

[Om Lydboblers Ydeevne: En Analyse af Retningsafhængig Præstation og Underliggende Akustiske Principper: En Analyse af Retningsafhængig Præstation og Underliggende Akustiske Principper]