Combining Deep and Unsupervised Features for Multilingual Speech Emotion Recognition

In this paper we present a Convolutional Neural Network for multilingual emotion recognition from spoken sentences. The purpose of work was to build model capable recognising emotions combining textual and acoustic information compatible with multiple languages. derive has an end-to-end deep architecture, hence it takes raw text audio data uses convolutional layers extract hierarchy classification features. Moreover, show how the trained achieves good performances in different languages thanks usage unsupervised As additional remark, is worth mention that our solution does not require be word- or phoneme-aligned. proposed model, PATHOSnet, evaluated on corpora (IEMOCAP, EmoFilm, SES AESI). Before training, tuned hyper-parameters solely IEMOCAP corpus, which offers realistic recording transcription sentences emotional content English. final turned out provide state-of-the-art some selected sets four considered emotions.