Theme 5

This weeks theme is design research. I have not really heard about design research before this course I think. At least not to the same extent as qualitative and quantitative methods and such. I think the paper by Haibo Li et al. is really interesting! Especially since I am in to mobile development and football is also one of my big interests. I have never really thought of the vibration function in mobile phones as something else than just a way of getting notified of new calls and text messages. But on most mobile phones you can set different vibration patterns for calls and for text messages. So why not use it for something even more advanced, like this. I do not really get why they say that it is unlikely that users are willing to stream an entire game of football. For me when I watch a game, I would like to see it as a whole.

I think that prototypes play an important role in research. It is when you do a prototype that you really can let users try out your design, and it is easier for them to get an understanding of the concept you are developing.  I think it could be necessary to develop a proof of concept prototype if you want to see what issues or difficulties your idea might have.


For the “design research” paper I chose “Automatic speech recognition in cocktail-party situations: A specific training for separated speech” which was published in Journal of the Acoustical Society of America (Volume 131, Issue 2, pp. 1529-1535 (2012)). In the article the authors give an introduction to ASR (Automatic Speech Recognition) and says that quite much research has been made on the subject, but that is primarily on single source speech, and not in a cocktail-party setting. The authors use different diagrams to show the set-ups for their experiments and how they make calculations on their data. They describe what mathematical methods they are using and they also give detailed data on all their measurements. I am not familiar with all the mathematical methods used in this study, but I have heard of some of them during the Spectral Transforms course I took three years ago.

In this paper a framework for robust ASR in cocktail-party situations has been presented. This framework is based on a robust transformed model constructed from separated speech in diverse acoustic environments. Three experiments are conducted in this study: Single speech recognition, Speech recognition with interfering speech, and Simultaneous speech recognition. The experiments were conducted in a room with the measures 5 x 4 x 3 meters. For all three experiments the word recognition rate (WRR) is shown in percent as a function of the wall reflection factor in a diagram. The results of the experiments are that both source separation and specific training provide a considerable improvement in WRR (up to a 35%), reducing the existing mismatch between the training and test data.