1.
I’ve chosen the article Recognition of the Main Melody in a Polyphonic Symbolic Score using Perceptual Knowledge written by Anders Friberg and Sven Ahlbäck, and published in the Journal of new music research.
The study aims to see if they can predict the main melody in a polyphonic score, which is in this case is a MIDI-representation of a song. They begin by explaining the concept on melody and more specifically the main melody, and how this can be perceived. They then try to identify the features that they think will represent how the main melody is perceived in the most adequate way. They use a total of 12 features, some examples of those being pitch and articulation.
The method used for predicting the main melody is a multiple regression. What this means is that they are trying to find the best possible model for predicting the main melody based on their perceptual knowledge. The regression model allows them to input and analyze huge amounts of data automatically. Since this is done using machine learning they have the advantage of training their model, and therefore optimizing it before trying to perform the actual prediction.
One limitation using a regression model for prediction is the choice of features may not always correctly correspond to reality, and I guess this is hard to actually prove. A more specific problem with this particular study might be the fact that they use a symbolic representation of music for analysis, while their perceptual knowledge comes from analyzing actual music, and due to this the result might a bit compromised.
I’ve chosen the article Recognition of the Main Melody in a Polyphonic Symbolic Score using Perceptual Knowledge written by Anders Friberg and Sven Ahlbäck, and published in the Journal of new music research.
The study aims to see if they can predict the main melody in a polyphonic score, which is in this case is a MIDI-representation of a song. They begin by explaining the concept on melody and more specifically the main melody, and how this can be perceived. They then try to identify the features that they think will represent how the main melody is perceived in the most adequate way. They use a total of 12 features, some examples of those being pitch and articulation.
The method used for predicting the main melody is a multiple regression. What this means is that they are trying to find the best possible model for predicting the main melody based on their perceptual knowledge. The regression model allows them to input and analyze huge amounts of data automatically. Since this is done using machine learning they have the advantage of training their model, and therefore optimizing it before trying to perform the actual prediction.
One limitation using a regression model for prediction is the choice of features may not always correctly correspond to reality, and I guess this is hard to actually prove. A more specific problem with this particular study might be the fact that they use a symbolic representation of music for analysis, while their perceptual knowledge comes from analyzing actual music, and due to this the result might a bit compromised.
2.
I think that the paper shows the complexity of performing a large quantitative study. There are so many variables to take into consideration and in this paper there are several examples of this.
First you have you have to gather participants, and in this case they sent out 5000 emails where 1509 went through to follow up questions and 74% of these participated throughout the whole study, which I guess is a good response rate. Furthermore there are a lot of parameters outside the study that could potentially affect the result, such as pollen season and influenza season, but also how to actually separate influenza from a common cold. I think that as long as you are aware of what the implications might be when performing a study like this, you can always bring that up for discussion in the end, which is done with influenza and a common cold.
So what have I learned? I think that it is of great importance that you motivate and explain your choice of method, questions and participants. If this is not done correctly (not saying that it’s not in the paper) the result of the actual study would probably be mostly speculations.
Which are the benefits and limitations of using quantitative/qualitative methods?
I think the obvious benefit is that it enables you to analyze a large set of data which I think gives you room to draw a bit broader conclusions, whereas a qualitative study might not. On the other hand the benefit of performing a qualitative study is that you might receive insight on a deeper level than you could using a quantitative method. Of course this depends on what you are investigating or writing about. If you have a hypothesis, you might want to try it using a quantitative method, and if the result differs from your expectation, you could further investigate why this is using a qualitative method. So perhaps you could say that the limitation of one the methods is the benefits of the other.
I see that the things you learned from the paper were not necessarily about the methods themselves, but about aspects connected to them such as the importance of motivating your choices which I find refreshing. I think that the previous themes of this course have been a little abstract and this is the first time where the theme is not completely foreign. Therefore the basics about methods probably doesn't come as news to any of us media technology students. I also agree that defining and motivating are the most important aspects of writing a research paper in order to get the best results possible!
SvaraRadera