Effects of Degradations on Internet Speech: Study on Quality Perception and Physiology, Slides of Computer Science

Download Effects of Degradations on Internet Speech: Study on Quality Perception and Physiology and more Slides Computer Science in PDF only on Docsity! The Good, the Bad and the Muffled: the Impact of Different Degradations on Internet Speech Docsity.com Introduction • Multimedia conferencing is a growing area • Well-known that need good quality audio for conferencing to be successful • Much research focused on improving delay, jitter, loss • Many think bandwidth will cure all problems – But bandwidth has been increasing exponentially while quality has not! Docsity.com Problems Cited • Missing words – Likely causes: packet loss, poor speech detection, machine glitches • Variation in volume – Likely causes: insufficient volume settings (mixer), poor headset quality • Variation in quality among participants – Likely causes: high background noise, poor headset quality • Experiments to measure which affect quality Docsity.com Outline Introduction ¢ Experiments Results * Conclusions ® Docsity.com Audioconference Fixed Parameters • Robust Audio Tool (RAT) – Home brewed in UCL – Does some repair of packets lost • Coded in DVI • 40 ms sample size • Use “repetition” to repair lost packets – Good for small packets (20ms) – Not as good for large packets (80ms) • (Why not?) Docsity.com Measurement Method: Physiological • User “cost” – Fatigue, discomfort, physical strain • Measure user stress – Using a sensor on the finger • Blood Volume Pulse (BVP) – Decreases under stress • Heart Rate (HR) – Increases under stress (“Fight” or “Flight”) Docsity.com Experimental Material • Take script from “real” audioconference • Act-out by two males without regional accents • Actors on Sun Ultra workstations on a LAN – Only audio recorded – 16 bit samples (DVI compresses to 4 bits) – Used RAT – Used silence deletion (hey, project 1!) • Vary volume and feedback (speakers to mic) • Split into 2-minute files, 8Khz, 40 ms packets • Repetition when loss Docsity.com Experimental Conditions • Reference – non-degraded • 5% loss – both voices, with repetition • 20% loss – both voices, with repetition • Echo – one had open mic, no headset • Quiet – one recorded low volume, other norm • Loud – one recorded high volume, other norm • Bad mic – one had low quality mic, other norm Determined “Intelligibility” not affected by above Docsity.com Outline Introduction ¢ Experiments Results * Conclusions ® Docsity.com Quality Under Degradation Bist O2nd | | ERR H Rg» LOO HH FSS QAQWQVAY RRA Ht Ss MMS’ Ho HH RSS REI. WWF WW oa oa ao oa So ao o a ao co bh ita} ira} = oo Cs _— Bune Aen’ ueayy Docsity.com acho loud 20% loss bad mic Degradation type ¢ Statistically significant? quiet feference 5% loss Statistical Significance Tests • ANOVA Test – For comparing means of two+ groups: first hearing and second hearing – No statistical difference between the two groups • Analysis of variance – Degradation effect significant • Reference and mean of all others are different – Reference and 5% loss the same – Reference and Quiet the same – 5% Loss and Quiet the same – 20% Loss and Echo and Loud the same Docsity.com Physiological Statistical Significance Tests • Bad mic, loud and 20% loss all significantly more stressful than quiet and 5% loss • Echo significantly more stressful than quiet in the HR data only • Contrast to quality! – Bad Mic worse than 20% loss – Least stressful were quiet and 5% loss Docsity.com Qualitative Results • Asked subjects to describe why each rating • Could clearly identify – quiet, loud and echo • Bad mic – ‘distant’, ‘far away’ or ‘muffled’ – ‘on the telephone’, ‘walkie-talkie’ or ‘in a box’ Docsity.com Qualitative Results of Loss • 5% loss – ‘fuzzy’ and ‘buzzy’ (13 of 24 times) • From waveform changing in the missing packet and not being in the repeated packet – ‘robotic’, ‘metallic’, ‘electronic’ (7 times) • 20% loss – ‘robotic’, ‘metallic’, ‘digital’, ‘electronic’ (15 times) – ‘broken up’ and ‘cutting out’ (10 times) – ‘fuzzy’ and ‘buzzy’ infrequently (2 times) • 5 said ‘echo’, 10 described major volume changes – Not able to reliably see the cause of the degradation Docsity.com