Dance emotion recognition based on linear predictive Meir frequency cepstrum coefficient and bidirectional long...

Area of Science:

• Robotics and Human-Computer Interaction
• Speech and Audio Processing
• Machine Learning for Affective Computing

Background:

• Dance emotion recognition is crucial for human-robot interaction, but faces challenges with small, high-dimensional datasets.
• Traditional methods like Recurrent Neural Networks (RNNs) suffer from vanishing gradients, while Convolutional Neural Networks (CNNs) struggle with long-range dependencies.
• Existing approaches often lack robustness and generalization in complex dance environments.

Purpose of the Study:

• To propose a novel feature extraction and classification framework for improved dance emotion recognition.
• To address the limitations of traditional Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) in handling dance emotion data.
• To enhance the robustness and generalization capabilities of emotion recognition systems in robotic applications.

Main Methods:

• A new feature, Linear Prediction Meier Frequency Cepstrum Coefficient (LPMFCC), is proposed by combining Linear Prediction Coefficients (LPC) and Meier Frequency Cepstrum Coefficients (MFCC).
• The LPMFCC feature is further combined with energy features for comprehensive dance feature extraction.
• A bidirectional Long Short-Term Memory (LSTM) network is employed for training these extracted features, followed by Support Vector Machine (SVM) classification.

Main Results:

• The proposed method, utilizing LPMFCC and bidirectional LSTM, demonstrated superior effectiveness compared to existing state-of-the-art dance motion recognition techniques.
• Experiments on public datasets validated the robustness and generalization of the developed acoustic model.
• The integration of novel features and advanced neural network architecture significantly improved dance emotion recognition accuracy.

Conclusions:

• The novel LPMFCC feature combined with bidirectional LSTM offers a promising solution for accurate and robust dance emotion recognition.
• This approach effectively overcomes the limitations of traditional methods in handling complex, high-dimensional dance emotion datasets.
• The findings contribute to advancing automatic speech recognition in robotic environments, enabling more nuanced human-robot interaction.