La comunicazione è uno scambio tra due individui, l’emittente invia segnali per influenzare il comportamento del ricevente. Il canto degli uccelli è stato ben studiato ed è noto che alcune variabili che determinano le differenze tra specie sono le dimensioni corporee, i fattori genetici e ambientali. Queste conoscenze non sono trasferibili tout court su altri taxa, come ad esempio i primati. Il canto è un tipo di comunicazione usato dai primati per difendere il territorio e in ambito riproduttivo. Le canzoni sono formate da elementi gerarchicamente ordinati e differiscono dalle vocalizzazioni per la durata, la struttura complessa e per l’incorporazione di unità vocali discrete. Le canzoni possono provenire da un solo individuo, o possono essere duetti o cori. I partecipanti possono essere dello stesso sesso o del sesso opposto, anche i giovani partecipano, probabilmente per imparare e fare pratica. Il canto è osservato in poche specie di primati, le famiglie coinvolte sono: Indriidae, Tarsiidae, Pitheciidae, Hylobatidae. Sono famiglie distanti tra loro, ma sono accomunate dalla strategia riproduttiva: la maggior parte delle specie che cantano attua la monogamia sociale, ma può coesistere con diverse altre strategie riproduttive. L’obiettivo di questa tesi è indagare le differenze temporali e spettrali tra le unità di canto dei primati e cercare di capire quali variabili possono giocare un ruolo nel determinare le eventuali differenze. In questa tesi sono state analizzate le caratteristiche temporali e spettrali della frequenza fondamentale (f0) di 254147 unità del canto di H. lar, I. indri, S. syndactylus. Tre algoritmi di machine learning (ML) sono stati utilizzati per assegnare le emissioni alla specie emittente ed è stata indagata la loro performance nel classificarle correttamente. È stata anche quantificata la sovrapposizione delle componenti principali delle unità del canto nell’ipervolume. Sono stati costruiti dei cluster gerarchici di distanza per ogni variabile la cui topologia è stata mappata sull’albero filogenetico delle tre specie. I risultati dei test non parametrici usati suggeriscono che ci sono differenze statisticamente significative in tutte le caratteristiche della f0 analizzate, I. indri ha la maggiore durata, la minor inclinazione dell’unità in Hz/s, e la più alta frequenza iniziale; H. lar ha la più ripida inclinazione dell’unità in Hz/s e la deviazione standard maggiore; S. syndactylus ha la minore durata, la minor deviazione standard, la minor f0 iniziale, la minor differenza tra la f0 finale e iniziale dell’unità. H. lar occupa il maggior spazio nell’ipervolume e ha la maggior variabilità, S. syndactylus occupa meno spazio, ha minor sovrapposizione con le altre specie e ha la minor variabilità, I. indri risulta intermedio. Le specie formano tre gruppi ben distinti. La alta percentuale di record classificati correttamente (98-99%) da tutti gli algoritmi di ML è coerente con le differenze tra specie per tutte le caratteristiche analizzate. La migliore performance è attribuita al multilayer perceptron (MLP). C’è correlazione tra la frequenza degli errori commessi dagli algoritmi di ML e la percentuale di sovrapposizione tra specie nell’ipervolume. Mediamente le variabili più importanti per la classificazione sono in ordine: f0 iniziale, durata, deviazione standard della f0, differenza tra la f0 finale ed iniziale divisa per la durata, differenza tra la f0 finale ed iniziale. [...]
Communication is an exchange between two individuals; the sender sends signals to influence the receiver's behaviour. Bird song has been well studied, and it is known that some variables that determine differences between species are body size and genetic, social, and environmental factors. This knowledge is not transferable to other taxa, such as primates. Singing is a type of communication primates use to defend their territory and in a reproductive context. Songs consist of hierarchically ordered elements and differ from vocalisations in duration, complex structure and the incorporation of discrete vocal units. Songs may originate from a single individual or be duets or choruses. Participants may be of the same or opposite sex, and youngsters may participate, probably to learn and practise. Singing is observed in a few primate species; the families involved are Indriidae, Tarsiidae, Pitheciidae, and Hylobatidae. These families are distant from each other but are united by their reproductive strategy: most singing species implement social monogamy but can coexist with various other reproductive strategies. This thesis aims to investigate the temporal and spectral differences between primate singing units and to try to understand which variables may play a role in determining any differences. This thesis analysed the temporal and spectral characteristics of the fundamental frequency (f0) of 254147 song units of H. lar, I. indri, S. syndactylus. Three machine learning (ML) algorithms were used to assign the emissions to the emitting species, and their performance in correctly classifying them was investigated. The overlap of the principal components of the song units in the hypervolume was also quantified. Hierarchical distance clusters were constructed for each variable whose topology was mapped onto the phylogenetic tree of the three species. The results of the non-parametric tests used suggest that there are statistically significant differences in all f0 characteristics analysed; I. indri has the longest duration, the smallest unit slope in Hz/s, and the highest initial frequency; H. lar has the steepest unit slope in Hz/s and the largest standard deviation; S. syndactylus has the shortest duration, the smallest standard deviation, the smallest initial f0, and the smallest difference between final and initial unit f0. H. lar occupies the greatest space in the hypervolume and has the greatest variability; S. syndactylus occupies the least space, has the least overlap with the other species and has the least variability; I. indri is intermediate. The species form three distinct groups. The high percentage of records classified correctly (98-99%) by all ML algorithms is consistent with the differences between species for all analysed features. The best performance is attributed to the multilayer perceptron (MLP). There is a correlation between the frequency of errors made by the ML algorithms and the percentage of overlap between species in the hypervolume. On average, the most important variables for classification are in order: initial f0, duration, the standard deviation of f0, the difference between final and initial f0 divided by duration, and the difference between final and initial f0. Duration and initial f0 show the same topology as the phylogenetic tree, but insufficient evidence exists to correlate them with phylogeny. [...]
Differenze temporali e spettrali tra le unità del canto dei primati
PORTAS, MANUELA
2023/2024
Abstract
Communication is an exchange between two individuals; the sender sends signals to influence the receiver's behaviour. Bird song has been well studied, and it is known that some variables that determine differences between species are body size and genetic, social, and environmental factors. This knowledge is not transferable to other taxa, such as primates. Singing is a type of communication primates use to defend their territory and in a reproductive context. Songs consist of hierarchically ordered elements and differ from vocalisations in duration, complex structure and the incorporation of discrete vocal units. Songs may originate from a single individual or be duets or choruses. Participants may be of the same or opposite sex, and youngsters may participate, probably to learn and practise. Singing is observed in a few primate species; the families involved are Indriidae, Tarsiidae, Pitheciidae, and Hylobatidae. These families are distant from each other but are united by their reproductive strategy: most singing species implement social monogamy but can coexist with various other reproductive strategies. This thesis aims to investigate the temporal and spectral differences between primate singing units and to try to understand which variables may play a role in determining any differences. This thesis analysed the temporal and spectral characteristics of the fundamental frequency (f0) of 254147 song units of H. lar, I. indri, S. syndactylus. Three machine learning (ML) algorithms were used to assign the emissions to the emitting species, and their performance in correctly classifying them was investigated. The overlap of the principal components of the song units in the hypervolume was also quantified. Hierarchical distance clusters were constructed for each variable whose topology was mapped onto the phylogenetic tree of the three species. The results of the non-parametric tests used suggest that there are statistically significant differences in all f0 characteristics analysed; I. indri has the longest duration, the smallest unit slope in Hz/s, and the highest initial frequency; H. lar has the steepest unit slope in Hz/s and the largest standard deviation; S. syndactylus has the shortest duration, the smallest standard deviation, the smallest initial f0, and the smallest difference between final and initial unit f0. H. lar occupies the greatest space in the hypervolume and has the greatest variability; S. syndactylus occupies the least space, has the least overlap with the other species and has the least variability; I. indri is intermediate. The species form three distinct groups. The high percentage of records classified correctly (98-99%) by all ML algorithms is consistent with the differences between species for all analysed features. The best performance is attributed to the multilayer perceptron (MLP). There is a correlation between the frequency of errors made by the ML algorithms and the percentage of overlap between species in the hypervolume. On average, the most important variables for classification are in order: initial f0, duration, the standard deviation of f0, the difference between final and initial f0 divided by duration, and the difference between final and initial f0. Duration and initial f0 show the same topology as the phylogenetic tree, but insufficient evidence exists to correlate them with phylogeny. [...]File | Dimensione | Formato | |
---|---|---|---|
801983_801983_manuela_portas_tesi.pdf
non disponibili
Tipologia:
Altro materiale allegato
Dimensione
3.26 MB
Formato
Adobe PDF
|
3.26 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14240/146133