Categorical and Dimensional Ratings of Emotional Speech: Behavioral Findings From the Morgan Emotional Speech Set

Abstract

Purpose Emotion classification for auditory stimuli typically employs 1 of 2 approaches (discrete categories or emotional dimensions). This work presents a new emotional speech set, compares these 2 classification methods for emotional speech stimuli, and emphasizes the need to consider the entire communication model (i.e., the talker, message, and listener) when studying auditory emotion portrayal and perception. Method Emotional speech from male and female talkers was evaluated using both categorical and dimensional rating methods. Ten young adult listeners (ages 19–28 years) evaluated stimuli recorded in 4 emotional speaking styles (Angry, Calm, Happy, and Sad). Talker and listener factors were examined for potential influences on emotional ratings using categorical and dimensional rating methods. Listeners rated stimuli by selecting an emotion category, rating the activation and pleasantness, and indicating goodness of category fit. Results Discrete ratings were generally consistent with dimensional ratings for speech, with accuracy for emotion recognition well above chance. As stimuli approached dimensional extremes of activation and pleasantness, listeners were more confident in their category selection, indicative of a hybrid approach to emotion classification. Female talkers were rated as more activated than male talkers, and female listeners gave higher ratings of activation compared to male listeners, confirming gender differences in emotion perception. Conclusion A hybrid model for auditory emotion classification is supported by the data. Talker and listener factors, such as gender, were found to impact the ratings of emotional speech and must be considered alongside stimulus factors in the design of future studies of emotion.

Effects of Increasing the Overall Level or Fitting Hearing Aids on Emotional Responses to Sounds

“…One potential explanation is that participant gender confounds ratings of unpleasant stimuli; Picou (2016) did not control for gender between participant groups, with more females in the groups with normal hearing and more males in the group with hearing loss. Some investigators have found effects of gender on ratings of valence where females provide lower ratings of valence of unpleasant stimuli than did males (e.g., Grossman & Wood, 1993 ; Natale et al, 1983 ; Vrana & Rollock, 2002 ), although others have not confirmed this effect (e.g., Morgan, 2019 ). Specific to the IADS stimuli, effects of gender on ratings of valence have also been mixed.…”

Emotion and Word Recognition for Unprocessed and Vocoded Speech Stimuli

“…Stimuli • Speech materials came from the MESS (Morgan 2019). The full set contains 1800 stimuli recorded by six talkers (3 male, 3 female) in four emotional speaking styles (Angry, Calm, Happy, and Sad).…”

“…To create these subsets of stimuli, all of the sentences from the female talkers in the MESS were sorted by their emotion recognition scores from a validation task (Morgan 2019), and the stimuli from each emotion category with the highest recognition rates were equally divided into nine sets. Pilot testing confirmed that these sets of 40 stimuli (10 stimuli × 4 emotions) from the MESS that had similar emotion recognition scores in quiet.…”

Improved remote mental health illness assessment and detection using facial emotion detection and speech emotion detection

“…There are multiple models out there for multi-dimensional representation of emotion such as the big five personality theory also known as the five-factor model proposed in [40]which classifies emotion into extroversion, agreeableness, neuroticism, conscientiousness, and openness to new experiences. Morgan [41]proposed the multi-dimensional representation for emotions which are calm, happy, sad, and angry. Two of the proposed dimensions were Activation level and pleasantness level.…”

“…Two of the proposed dimensions were Activation level and pleasantness level. But these representations proposed in [41]are limited to only four emotions. Mancini [42]provided a more diverse classification of emotions into valence and arousal, and we decided to proceed with those values.…”