Comparative study of features extraction for speech's emotion at micro and macro level

the presented paper is concerned with emotion recognition from speech based on micro and macro level features. Prosodic and temporal features are explored to identify their significance and contribution in emotion recognition system. Emotions are extracted at segment (micro) level and macro (utterance) level. The method has been verified using two emotional speech database with support vector machine classifier. Results at both levels are compared and better recognition rate are achieved at micro level than global statistics.