SEMOUR DATABASE – v1.0
The Scripted Emotional Speech Repository for Urdu (SEMOUR) is an acted, multimodal and multispeaker database, recently collected at Acoustics Lab lab at ITU, Lahore. This gender-balanced dataset contains 15, 040 unique instances and about 7 hours of recorded data by eight actors eliciting a syntactically complex script. In the current version, one-third of SEMOUR database instances are annotated by two annotators each into categorical labels, such as anger, happiness, sadness, neutrality, as well as dimensional labels such as valence, activation, and dominance. The detailed speech capture information, the interactive setting to elicit authentic emotions, and the size of the database make this corpus a valuable addition to the existing databases in the community especially for a low-resourced language like Urdu for the study and modeling of multimodal and expressive human communication.
Scope of the dataset
Recognition and Analysis of Emotional Expression
Design of Emotion-Sensitive Human-Computer Interfaces and Virtual Agents
Personalized Robot Assistants
Diagnosis of Psychological Disorders
Get Feedback from Low-tech-enabled Population
8 Actors: 4 male and 4 female
Emotion Elicitation Techniques: Script consisting of 235 instances common
Emotions: Anger, Boredom, Disgust, Fearful, Happiness, Neutral, Sadness, Surprise
The SEMOUR v1.0 database contains 235 instances uttered by 8 actors in 8 different emotions. The SEMOUR script is created keeping phoneme balance and common Urdu language words in mind. The script contains 43 daily life words including the name of colors, counting, etc. The 66 two-word phrases are where we maintain the phoneme balance in the script including somewhat complex words. The rest of the 126 sentences include emotionally charges text for all the 8 elicited emotions of SEMOUR. Urdu, with 36 alphabets, is a much more diverse language having more than one and a half times the phonemes as the English language with 26 alphabets. A phonetically balanced script with a copious depth of instances provides a better chance to cater to speech-related problems. The collected dataset has wide applications including speech-to-text, speaker identification, and speech emotion recognition, etc.
All the actors were volunteers contacted through social media advertisement. They were locales of province Punjab, Pakistan and had Urdu dialect. All of them had either Urdu or Punjabi accent. Further demographic details are provided for each actor in the table below.
|Actor ID||Gender||Age||Occupation||Home Town|
To download our sample dataset click here