Analiza intonacije petja z algoritmi za detekcijo višine tona

• Main Author: ŠIŠKOVIČ, MARTIN
• Other Authors: Rozman, Robert
• Format: Bachelor Thesis
• Language: Slovenian
• Published: 2017
• Subjects: algoritmi za detekcijo višine tona; analiza petja; avtokorelacija; intonacija; sprememba intonacije; pitch detection algorithms; singing analysis; autocorrelation; intonation; pitch drift; sami
• Online Access: https://repozitorij.uni-lj.si/IzpisGradiva.php?id=91342; https://repozitorij.uni-lj.si/Dokument.php?id=97323&dn=

Splošno znan pojav pri pevcih je spreminjanje intonacije med petjem. Z njim se srečujejo vsi, ki se ukvarjajo s petjem brez instrumentalne spremljave. Kljub njegovi pogostosti in nezaželenosti ni znan enoten razlog za prisotnost pojava. Običajno se zmotno trdi, da je spreminjanje intonacije odvisno od kvalitete pevcev. Cilj diplomske naloge je poiskati vzroke za spremembe v intonaciji. Narejen je bil pregled znanstvenih objav, ki so povezane s tematiko ter analiza sprememb v intonaciji na realnih posnetkih. Posnetke smo sami naredili v okviru diplomske naloge. Ti so bili analizirani z algoritmi za detekcijo višine tona. Uporabili smo obstoječi vtičnik algoritma pYIN in implementirano lastno različico avtokorelacije. Rezultati analize so dodatno potrdili ugotovitve nekaterih objav, ki pravijo, da se intonacija v skladbi spreminja zaradi razlik med naravno in temperirano uglasitvijo. Spremembe se pojavijo, ko je med dvema tonoma razlika med uglasitvama negativna. Več kot je takih primerov v skladbi, večja je lahko sprememba v intonaciji. Pitch drift is a commonly known phenomenon among singers. It occurs in most a cappella groups. Reasons for the phenomenon are not known, despite that it is very common and unwanted. It is usually mistakenly argued that pitch drift depends on the singers skills. The objective of this thesis is to find the causes of pitch drift. We reviewed scientific publications related to the theme and analysed the changes in intonation on audio tracks, which we recorded in the context of this thesis. Tracks were analyzed with algorithms for pitch detection. We used an existing plug-in based on pYIN algorithm and our own modified implementation of algorithm based on autocorrelation. The analysis results have further confirmed the findings of some publications, saying that the intonation in the songs changes due to differences between the just intonation and the equal temperament. Changes occur when difference in tuning between two notes is negative. More such cases are there in the song, greater change in the intonation can occur.