Emotional Corpus, Feature Extraction and Emotion Classification Using the Parameterized Voice Signal

The recognition and classification of human emotions through voice analysis, it is a very interesting research area, due to the wide variety of applications: telecommunications, learning, human-computer interface, entertainment, etc. In this investigation a methodology is proposed for the recognition of emotions analyzing voice segments. The methodology is mainly based on the fast Fourier transform (FFT) and Pearson's correlation coefficients. The tone (pitch), the fundamental frequency (Fo), the strength of the voice signal (energy) and the speech rate have been identified as important indicators of the emotion in the voice. The system consist of a graphical interface that allows user interaction by means of a microphone integrated into the computer, which automatically processes the data acquired. In our environment, human beings are programmed to let our voice flow, in multiple ways to communicate and to capture through it emotional states. There are various investigations where the Berlin database is used, which is free and many researchers have used it in their research. However, the creation of an emotional corpus with Spanish phrases, was needed for testing that provide clearer results. The corpus contains 16 phrases per emotion created by 11 users (9 women and 2 men) with a total of 880 audio samples. The following basic emotions were considered: disgust, anger, happiness, fear and neutral. Results obtained indicate that the emotion recognition algorithm offers an 80% of effectiveness.