Analysis of sound for emotion speech recognition

Abstract

Recognising emotion from the speech or Emotion Speech Recognition has been relatively recent research field in the speech recognition. This is useful for applications which require natural interaction between human and machine, for example ticket reservation machine, call centre application, as well as in medical field. However, getting the reliable model for Emotion Speech Recognition is a challenge. In this report, four basics emotion (e.g. Happy, Angry, Anxious, and Sad) will be used to analyse the best features and classification model for Emotion Speech Recognition. Different approaches are explored and analysed to determine which approach gives the highest accuracy, using WEKA as the classification tool and Praat to extract the speech features. After experimenting with Praat, the best approach is implemented into real-time mobile application with Android platform. TarsosDSP is used as the external library to process the audio signal, as well as extract the speech features needed.