Recognizing Emotional States Using Speech Information.

Area of Science:

• Speech processing
• Machine learning
• Affective computing

Background:

• Emotion recognition is crucial for human-computer interaction and monitoring user states.
• Voice is a key modality for emotion recognition, especially when other data is unavailable.
• Existing methods often rely on linguistic content, but paralinguistic cues offer a distinct avenue.

Purpose of the Study:

• To analyze speaker emotions using solely paralinguistic information from voice.
• To compare the effectiveness of Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs) for this task.
• To develop a multilingual emotion recognition system.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) trained on raw speech data.
• Employed Support Vector Machines (SVMs) trained on extracted low-level acoustic features.
• Compiled multilingual datasets for training and testing to ensure broad applicability.

Main Results:

• Both CNN and SVM models demonstrated capability in recognizing emotions from speech.
• Paralinguistic analysis proved effective for emotion detection across different languages.
• The study provides a comparative analysis of two distinct machine learning approaches.

Conclusions:

• Voice-based emotion recognition using paralinguistic features is feasible with current machine learning techniques.
• CNNs and SVMs offer viable, albeit different, pathways for developing such systems.
• The proposed multilingual approach enhances the potential for real-world applications.