University of Ljubljana, 2012. - 116 p.
The two main objectives of this project are to analyse the efficiency of several techniques widely used among the field of emotion recognition through spoken audio signals, and, secondly, obtain empirical data that proves that it is actually plausible to do so with a more than acceptable performance rate.
For that purpose, our research will follow four main stages. First, we will board the theoretical approach behind affective states as well as their classification. After that, we will introduce the relationship between spoken signals and their intrinsic acoustic features, through which it is possible to decode information about emotion states. The second stage will cover the presentation of our available materials and chosen methodology. To do so, a description of our source of speech files – emotion database – , as well as a thorough analysis of the most widely used acoustic features meant to decode the hidden information about emotions will be covered. The final part of this stage will treat several feature selection and classification techniques, as well as the means through which we will be presented with the final data.
The third stage, we will cover the presentation of the selected tools that will make possible the whole experimentation part: the feature extraction tool, and the data-mining software that will deliver the final classified data. The fourth and final stage will contain all our experimental work and the final conclusions. Here, we will work with different sets of acoustic features, we will classify them under different nature classifiers and analyse the results. Even more, we will apply several feature selection techniques to these extracted attribute sets, classify them and compare their performance. Each experiment will have its own conclusions, but there will also be one last chapter of general discussion that will take into account all the experimentation.