Impact of Feature Selection Algorithm on Speech Emotion Recognition Using Deep Convolutional Neural Network.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Speech Processing

Background:

• Speech emotion recognition (SER) is crucial for human-machine interaction but challenging due to context limitations.
• Traditional handcrafted features are insufficient for accurately capturing speaker emotional states.
• Deep learning offers potential for improved feature extraction in SER.

Purpose of the Study:

• To explore the benefits of deep convolutional neural networks (DCNNs) for speech emotion recognition.
• To extract emotionally relevant features using a pretrained DCNN.
• To evaluate the effectiveness of DCNN-extracted features compared to handcrafted features.

Main Methods:

• Utilized a pretrained DCNN for feature extraction from speech emotion datasets.
• Applied a correlation-based feature selection technique to identify discriminative features.
• Employed classifiers including SVM, Random Forests, k-NN, and neural networks for emotion classification.

Main Results:

• Achieved high accuracy in speaker-dependent SER: 95.10% (Emo-DB), 82.10% (SAVEE), 83.80% (IEMOCAP), 81.30% (RAVDESS).
• Demonstrated superior performance for speaker-independent SER compared to existing handcrafted feature methods.
• Validated the effectiveness of DCNN-based feature extraction for SER tasks.

Conclusions:

• DCNNs significantly improve feature extraction for speech emotion recognition.
• The proposed method offers a robust approach for both speaker-dependent and independent SER.
• This research advances the field of affective computing and human-computer interaction.