1) Tell us a little about yourself. Where are you from and what do you study?
I'm from Northern Poland, however I spend most of the time in Cracow right now. The reason is included in the second part of the question - I study Acoustic Engineering at the AGH University of Science and Technology, finishing my 2nd cycle studies soon.
2) What initiated your passion for audio? When did it start?
I think the first time I thought about becoming a professional in acoustics or audio was in high school. In those times I was really interested in physics generally and after starting thinking about my future I found that sound is one of the most interesting phenomena for me. It spoke to me both scientifically as a purely physical process and emotionally as I've always loved listening to music and any other sounds produced by nature, which made me curious what's the real process behind that.
3) Tell us about production of your submission? What is the story behind it? How long did you work on it? Was it your first entry? What kind of problem can it solve or improve?
My project is entitled "Spectral Contrast Based Feature Extraction Algorithm for Acoustic Event Classification". The goal of the project was to create an efficient algorithm, which would extract a proper representation of a signal for several acoustic event classes. This is one of the crucial parts of a machine learning pipeline, in which we're training models (for example Neural Networks or Gaussian Mixture Models) with some feature vectors. This way our system is learning to recognize some specific types of signals in which we're interested in. In this case I've created an algorithm for acoustic event classification, so I needed something possibly general, as acoustic events include a very wide range of acoustic signal types. Experiments showed really promising results, in which the new algorithm outperformed some other commonly used methods.
This thing is just a part of a wider research I started for my BSc thesis over a year ago. During my studies I've became specially interested in DSP algorithms and systems using machine learning methods. Doing some initial research I found that Acoustic Event Detection and Classification is a field still to be studied and improved and in my humble opinion it's one of the most promising audio topics to become fruitful in the future. It has many possible or already existing practical applications like surveillance systems, smart home decisions, acoustic scene analysis, cryminology or speech recognition enhancement. Naturally this conclusion altogether with my interests led me to step into this world and try to find my own place. So it's been over a year now as I'm doing a constant research in this field and I'm definitely going to continue it in the near future. Currently I'm reading a lot of papers and articles related with my MSc Thesis, in which I'm going to use deep neural networks. Pretty exciting.
4) Did you considered commercializing your project? Are there any business or product possibilities?
Not so much. Unfortunately the topics I usually work on are not easily commercializable ones, as they require many years of research and hard work to get anything near a proper product. As a future professional I'm rather thinking about joining a team enabling a profitable co-operation which would lead us to some point where we have a working system, ready to sell. That would be a dream-come-true compromise between my scientific and business aspirations.
5) Do you know or consider any future steps? Will it be linked with the project you’ve presented?
As I mentioned before, I continue my studies on acoustic event classification. For now I'm focused on preparing my MSc thesis related to the use of deep learning in this context. However I'm pretty sure it will be just a next step and many other are waiting to come. This shall be a long journey.
Posted: Monday, December 3, 2018