Quantification of similarity between dickcissel song dialects
(2020) In Master's Theses in Mathematical Sciences FMSM01 20201Mathematical Statistics
- Abstract
- In this thesis, five methods of scoring the similarity of phrases from dickcissel birds are explored. The methods are tested on real and simulated data. The method that is found to perform the best is a variation of the spectrographic cross-correlation method that also looks at correlation in the frequency domain. The second best method is based on singular value decomposition of the spectrogram of the phrase and extracting density-based features. While the methods are not good enough to give a reliable similarity between two phrases, they are able to find structure in a larger data set.
- Popular Abstract
- Using computers to analyze bird songs has given us great possibilities in recent times. Not only does it reduce the workload of researches looking through audio files for the presence of birds it also allows for tasks that would be impossible otherwise such as computing the pairwise similarity between thousands of songs or tracking the positions of birds in real time. One particularly interesting application tracks the locations of birds around airports to prevent collisions between birds and aircraft.
While big steps have been made in the past there are still many obstacles left and all known methods have some downside that makes them unsuitable for some tasks. What is often done these days is to develop a method specifically for the... (More) - Using computers to analyze bird songs has given us great possibilities in recent times. Not only does it reduce the workload of researches looking through audio files for the presence of birds it also allows for tasks that would be impossible otherwise such as computing the pairwise similarity between thousands of songs or tracking the positions of birds in real time. One particularly interesting application tracks the locations of birds around airports to prevent collisions between birds and aircraft.
While big steps have been made in the past there are still many obstacles left and all known methods have some downside that makes them unsuitable for some tasks. What is often done these days is to develop a method specifically for the bird species analyzed. Timothy H. Parker of Whtiman College, Walla Walla, has collected dickcissel songs and wants to find an algorithm that can compare two songs and say how similar they are. The goal is to see how bird culture changes over time and space and what the effect of different environments are on songs. In the thesis a method is found that is specialized towards the similarity of dickcissel songs.
Several methods are discussed and some are examined in greater detail. To test the methods some artificial songs are synthesized and by analyzing the way the methods react to different signals properties can be derived. For example some of the methods are very sensitive to shifts in the pitch of the signal while some are not. The methods are also evaluated on the real songs collected in Kansas, USA. Based on how accurate the methods are on these songs the performance can be evaluated.
The method that ends up performing the best is a well known general method called spectrographic cross-correlation. It works by seeing how well the signals correlate in time and frequency. Using this method some structure can be found in the data set but further analysis is needed to make statements about the birds themselves. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9025533
- author
- Andrén, Patrik LU
- supervisor
- organization
- course
- FMSM01 20201
- year
- 2020
- type
- H2 - Master's Degree (Two Years)
- subject
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUTFMS-3398-2020
- ISSN
- 1404-6342
- other publication id
- 2020:E68
- language
- English
- id
- 9025533
- date added to LUP
- 2020-10-05 12:56:40
- date last changed
- 2021-06-03 17:36:52
@misc{9025533, abstract = {{In this thesis, five methods of scoring the similarity of phrases from dickcissel birds are explored. The methods are tested on real and simulated data. The method that is found to perform the best is a variation of the spectrographic cross-correlation method that also looks at correlation in the frequency domain. The second best method is based on singular value decomposition of the spectrogram of the phrase and extracting density-based features. While the methods are not good enough to give a reliable similarity between two phrases, they are able to find structure in a larger data set.}}, author = {{Andrén, Patrik}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Theses in Mathematical Sciences}}, title = {{Quantification of similarity between dickcissel song dialects}}, year = {{2020}}, }