The Relevance of Voice Quality Features in Speaker Independent Emotion Recognition
IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. IV-17-IV-20
Bayes methods;emotion recognition;speaker recognition;speech processing;Bayesian classifier;phonation types;prosodic features;speaker independent;speaker independent emotion recognition;voice quality features;Bayesian methods;Emotion recognition;Feature extraction;Mel frequency cepstral coefficient;Pattern classification;Production;Psychology;Signal processing;Spatial databases;Speech analysis;Feature extraction;Pattern classification;Speech analysis
- Marko Lugger
- Bin Yang
This paper investigates the classification of different emotional states using presodic and voice quality information. We want to exploit the usage of different phonation types within the production of emotions. Therefore, as features we use prosodic features, voice quality parameters, and different combinations of both types. We study how prosodic and voice quality features overlap or complement each other in the application of emotion recognition. The classification is speaker independent and uses a reduced subset of 8 features and a Bayesian classifier.