Balancing Spoken Content Adaptation and Unit Length in theRecognition of Emotion and Interest

摘要

Recognition and detection of non-lexical or paralinguistic cues from speech usually uses one general model per event (emo-tional state, level of interest). Commonly this model is trained independent of the phonetic structure. Given sufficient data, this approach seemingly works well enough. Yet, this paper ad-dresses the question on which phonetic level there is the onset of emotions and level of interest. We therefore compare phoneme-, word- and sentence-level analysis for emotional sentence clas-sification by use of a large prosodic, spectral, and voice qual-ity feature space for SVM and MFCC for HMM/GMM. Exper-iments also take the necessity of ASR into account to select appropriate unit-models. In experiments on the well-known public EMO-DB database, and the SUSAS and AVIC sponta-neous interest corpora, we found that the emotion recognition by sentence level analysis shows the best results. We discuss the implications of these types of analysis on the design of ro-bust emotion and interest recognition of usable human-machine interfaces (HMI).