Zainab984/CS461-KESC-Anonymous-Dataset

CS461 KESC Anonymous Dataset

The KESC dataset contains anonymized short dramatic-scene records for research on emotion recognition in Kuwaiti Arabic. It includes Arabic dialogue transcriptions, English translations where available, standardized affective labels, and paired audio and video files.

Dataset contents

• 1,335 annotation records
• 1,335 records with paired MP4 video and MP3 audio
• 21 columns in annotations.xlsx
• Neutral record and media identifiers

Files

• annotations.xlsx contains the anonymized annotation table.
• videos/ contains video clips named with neutral KESC identifiers.
• audio_clips/ contains audio files paired with the released videos.
• KESC_Dataset_Description.docx provides the complete dataset description, schema, limitations, and release requirements.

Annotation fields

The workbook includes the sample identifier, video and audio paths, drama and episode information, scene timing and duration, Arabic and English dialogue, speaker labels, standardized emotion labels, expression type, polarity, intensity, face visibility, audio clarity, language, speaker gender, number of people, and bullying status.

Recommended use

The dataset can support speech, text, video, and multimodal emotion-recognition research. Train, validation, and test partitions should be grouped by drama and episode to reduce information leakage between related scenes.