Emotion identification using extremely low frequency components of speech feature contours.

Area of Science:

• Speech processing
• Machine learning
• Affective computing

Background:

• Emotional speech identification is crucial for human-computer interaction.
• Existing methods often rely on standard acoustic features, limiting performance.
• A need exists for more robust and informative speech features for emotion recognition.

Purpose of the Study:

• To develop an effective speech feature set for enhanced emotional speech identification.
• To integrate statistical acoustic features with approximated speech feature contours.
• To evaluate the proposed feature set using multiclass emotion identification.

Main Methods:

• Extracted frame-based acoustical features and approximated speech feature contours.
• Applied principal component analysis (PCA) to speech feature contours for efficient representation.
• Utilized support vector machines (SVMs) for multiclass emotion identification.

Main Results:

• The proposed speech feature set demonstrated superior performance in emotion identification.
• Achieved a high identification rate of 82.26% for multiclass emotion recognition.
• Principal component analysis effectively reduced dimensionality while preserving contour information.

Conclusions:

• The novel speech feature set significantly improves emotional speech identification accuracy.
• The integration of approximated contours and PCA offers a promising direction for affective computing.
• The SVM classifier effectively leverages the proposed features for robust emotion detection.