Recording and evaluation of speech data set for negative emotions recognition in speech


The article describes the approach of recording a database, based on a specially developed list of phrases. The list of emotions included 7 emotions, as the basic ones: disgust, sadness, contempt, joy, irritation, fear, surprise; the list also included the emotion of neutrality, and the emotion of smirk, which is a complex emotion and can be interpreted by people in different ways. The total list of phrases consisted of 40 utterances of different lengths, the total list was divided into 5 sublists of 8 phrases for each speaker. The speaker was presented with the lexical composition of the target phrase and its corresponding context, attitudes, i.e. emotions were removed for the clarity of the experiment. This approach to recording the dataset was chosen in order to obtain more natural realizations of emotion within the studio. After recording the speakers, a primary dataset was generated and validated with auditory experiment. Preliminarily to the experiment all the auditors were tested to determine the level of emotional intelligence according to the method of N. Hall.