UrduSER: A comprehensive dataset for speech emotion recognition in Urdu language.

Area of Science:

• Speech Emotion Recognition (SER)
• Computational Linguistics
• Machine Learning

Background:

• Speech Emotion Recognition (SER) is vital for understanding human-computer interaction and has significant socio-cultural and business applications.
• Existing datasets for Urdu, the 10th most spoken language, are limited in scope, emotion range, and dialogue diversity, hindering real-world SER applications.
• A significant research gap exists in Urdu SER due to the scarcity of comprehensive and diverse speech emotion datasets.

Purpose of the Study:

• To develop a comprehensive and balanced Speech Emotion Recognition Dataset for the Urdu language (UrduSER).
• To address the limitations of existing Urdu SER datasets by incorporating diverse, real-world dialogues and a wider range of emotions.
• To facilitate advancements in machine learning and deep learning models for Urdu speech emotion analysis.

Main Methods:

• Collected 3500 speech signals from 10 professional Pakistani actors (balanced gender and age) from YouTube drama serials and telefilms.
• Included seven distinct emotional states: Angry, Fear, Boredom, Disgust, Happy, Neutral, and Sad, with 500 samples per emotion.
• Ensured dialogue diversity with unique content per utterance and provided detailed metadata, including scripts, for each audio sample.

Main Results:

• Developed the UrduSER dataset, a comprehensive resource featuring 3500 diverse speech signals.
• The dataset includes a balanced distribution of 500 samples per emotion and 50 samples per actor per emotion.
• Expert validation confirmed the dataset's validity, reliability, and suitability for research and development.

Conclusions:

• The UrduSER dataset effectively fills the critical research gap for Urdu speech emotion recognition.
• Its diverse, real-world nature and comprehensive metadata enhance its utility for training robust SER models.
• This resource is expected to significantly advance research and development in Urdu SER, enabling more accurate and nuanced emotion detection in spoken language.