In:
The Journal of the Acoustical Society of America, Acoustical Society of America (ASA), Vol. 126, No. 4_Supplement ( 2009-10-01), p. 2221-2221
Abstract:
While discriminating between several basic human emotions, such as neutral, joy, sadness, and anger, it has been observed that the most difficult emotions to tell apart automatically are between the two states of joy and anger. Since these two emotions have similarities on the arousal scale, it is difficult to distinguish them by simply using pitch and energy related feature measurements. Therefore, in this study, other additional feature parameters, related to voice quality, that are useful for discriminating between the two emotions of joy and anger are focused. For voice quality related features, global statistics of normalized spectral band energy, spectral tilt, open quotient, and first formant bandwidth values, along with their respective slopes and convexities, are measured from the happy and angry emotional speech from a Korean emotional database. From ANOVAtests, parameters of normalized spectral band energy, spectral tilt, and open quotient appear to be useful. Also, slopes and convexities of voice quality measurements appear to be more important than the values itself. These results are meaningful for classifying emotional states distributed on the valence scale and are expected to contribute in improving an overall emotion recognition system.
Type of Medium:
Online Resource
ISSN:
0001-4966
,
1520-8524
Language:
English
Publisher:
Acoustical Society of America (ASA)
Publication Date:
2009
detail.hit.zdb_id:
1461063-2
Bookmarklink