Sustav za dijagnostiku i praćenje intenziteta mucanja primjenom strojnog učenja na rubnim uređajima

master's thesis

• Title (english): System for diagnosing and monitoring the intensity of stuttering using machine learning on edge devices
• Author: Petar Sušac
• Mentor: Hrvoje Džapo (mentor)
• Committee member: Hrvoje Džapo (predsjednik povjerenstva)
• Committee member: Mario Cifrek (član povjerenstva)
• Committee member: Željka Lučev Vasić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-16, Croatia

Abstract

Mucanje je poremećaj tečnosti govora koji u značajnoj mjeri utječe na verbalnu komunikaciju. Mjerenja intenziteta mucanja i anksioznosti povezane s govorom važna su za uspješno provođenje logopedske terapije, no zbog velike varijabilnosti intenziteta mucanja pouzdana mjerenja su nepraktična i zahtijevaju velik trošak vremena logopeda. U ovom radu razvijen je sustav za automatizirano mjerenje intenziteta mucanja i prikupljanje biomedicinskih signala koji ukazuju na anskioznost. Razvijen je model umjetne neuronske mreže koji detektira tri osnovne vrste mucajućih netečnosti u govoru: blokade, produljenja i ponavljanja. Razvijena je programska potpora za nosivi ugradbeni računalni sustav koji pomoću navedenog modela u stvarnom vremenu mjeri intenzitet mucanja na temelju govora, prikuplja signale vodljivosti kože i fotopletizmografije i iz njih izdvaja odabrane značajke te sprema rezultate obrade za kasniju analizu i vizualizaciju. Sustav postiže točnost do 71% u detekciji mucajućih netečnosti.

Abstract (english)

Stuttering is a speech fluency disorder with a significant effect on the verbal communication ability of those who stutter. Being able to measure the intensity of stuttering and the level of speech-related anxiety in patients is important to speech pathologists in order to successfully conduct speech therapy. However, due to the large variability of stuttering intensity, reliable measurements are impractical and time-consuming. This thesis presents the development of a wearable automated stuttering intensity and speech-related anxiety measurement system. An artificial neural network model is developed to detect the three types of stuttering disfluency (blocks, prolongations, repetitions) from speech, as well as software for a wearable embedded system which uses the neural network model to measure the stuttering intensity in real time. In addition, the wearable system extracts selected features of electrodermal activity and photoplethysmography biosignals which can be linked to the level of anxiety. The results are stored to non-volatile storage for subsequent analysis and visualization. The system achieves up to 71% accuracy in stuttering disfluency classification.

• Keywords: afektivno računarstvo; duboko učenje; mucanje; ugradbeni sustavi; Zephyr RTOS
• Keywords (english): affective computing; deep learning; stuttering; embedded systems; Zephyr RTOS
• Language: croatian
• URN:NBN: urn:nbn:hr:168:629033
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-10-12 22:11:05