Fast Individual HRTF Acquisition with Unconstrained Head Movements for 3D Audio
Author
Molina Garcia, Javier
Term
4. Term
Education
Publication year
2019
Submitted on
2019-05-28
Abstract
Hovedrelaterede overførselsfunktioner (HRTF) beskriver, hvordan hoved, ører og overkrop ændrer lyd, der kommer fra forskellige retninger. De gør det muligt for hovedtelefoner at genskabe oplevelsen af lyd, der omgiver lytteren (3D-lyd). Mange studier peger på, at individuelle HRTF’er er nødvendige for en overbevisende, immersiv oplevelse. Dette projekt implementerer en metode baseret på adaptiv filtrering, som gør det muligt for brugere at opnå personlige HRTF’er ud fra egne binaurale optagelser (med mikrofoner ved begge ører) kombineret med hovedsporing. Vi udførte simulationer for at undersøge, hvordan fire forhold påvirker nøjagtigheden: signal-støj-forholdet (hvor stærkt testsignalet er i forhold til baggrundsstøj), typen af excitationssignal, mønstre i hovedbevægelser under optagelsen og længden af indsamlingen. Resultaterne bekræfter metoden: systemet kan estimere HRTF’er præcist og opnår tilstrækkelig præcision selv ved meget lavt signal-støj-forhold (under 30 dB).
Head-related transfer functions (HRTFs) describe how the head, ears, and torso shape sounds arriving from different directions. They allow headphones to recreate the feeling of sound coming from around the listener (3D audio). Many studies indicate that individualized HRTFs are needed for a convincing, immersive experience. This project implements an adaptive filtering method that lets users obtain personalized HRTFs from their own binaural recordings (made with microphones at both ears) combined with head-tracking information. We ran simulations to examine how four factors affect accuracy: the signal-to-noise ratio (how strong the test sound is compared with background noise), the type of excitation signal, the pattern of head movements during recording, and the length of the acquisition. The results validate the approach: the system estimates HRTFs accurately and maintains adequate precision even when the signal-to-noise ratio is very low (below 30 dB).
[This abstract was generated with the help of AI]
Documents