Recognition of emotions in the Algerian dialect speech

Abstract

Emotion recognition from speech has been an important and challenging research. The performance of speech features and influence of emotions number on the systems of emotion recognition in the Algerian dialect speech are studied in this work. To achieve this aim, an emotional database of the Algerian dialect (ADED) is built. This database contains four emotions: anger, fear, sadness and neutral. Extraction of features is an important step in the recognition of speech emotion. The features extracted in this study are: pitch, intensity, duration, unvoiced frames, jitter, shimmer, HNR (Harmonic Noise Ratio), formants and MFCCs (Mel Frequency Cepstral Coefficients). Each recognition system needs a classifier, so our system is based on the KNN (K-Nearest Neighbor) method of classification. Different results are obtained, the higher recognition rates are given when using a combination of features including pitch, intensity, duration, unvoiced frames, jitter, shimmer, HNR and the MFCCs parameters. And the performance is influenced by the number of emotions included: 82.29 % when using the four emotions of ADED database, 84.02% when three emotions fear, anger and neutral are recognized and 87.50% when only the fear and neutral emotions are used in the recognition system.