Recognize basic emotional statesin speech by machine learning techniques using mel-frequency cepstral coefficient features

摘要: Speech Emotion Recognition (SER) has been widely used in many fields, such as smart home assistants commonly found the market. Smart that could detect user’s emotion would improve communication between a user and assistant enabling to offer more productive feedback. Thus, aim of this work is analyze emotional states speech propose suitable algorithm considering performance verses complexity for deployment devices. The four sets were selected from Berlin Emotional Database (EMO-DB) experimental data, 26 MFCC features extracted each type identify emotions happiness, anger, sadness neutrality. Then, speaker-independent experiments our conducted by using Back Propagation Neural Network (BPNN), Extreme Learning Machine (ELM), Probabilistic (PNN) Support Vector (SVM). Synthesizing recognition accuracy processing time, shows SVM was best among methods good candidate be deployed SER achieved an overall 92.4% while offering low computational requirements when training testing. We conclude classification models are highly effective automatic prediction emotion.