ADMRF: Elucidation of deep feature extraction and adaptive deep Markov random fields with improved heuristic algorithm for speech emotion recognition

On considering Human–Computer Interaction, the recognition of emotion over speech is developed from the niche phase into the most essential component. But, it is regarded as the most challenging issue due to the unclear features that are required to perform this task. In most of the conventional studies, various approaches have been carried out for retrieving the emotions from the signals for performing Speech Emotion Recognition that may involve diverse categorization and well-established speech analysis. Further, it has a better enhancement for upgrading human emotion analysis. This paper aims to build an intelligent recognition model using the improved heuristic algorithm with an adaptive deep learning approach. The foremost step is the collection of speech signals that are fetched from the standard data sources. It is then followed by the pre-processing phase, where the denoised signal is obtained. Then, the Short Term Fourier Transform is applied for generating the spectrogram image. Subsequently, the deep features are extracted through resultant image by influencing the method as an Attention-based Multi-scale Visual Geometry Group16 (AM-VGG16). Finally, the speech emotions are recognized by using the Adaptive Deep Markov Random Fields (ADMRF) with the help of attained deep features. Further, the parameter tuning is take place in the ADMRF by proposing the new algorithm as Modified Position in Squid Game Optimizer for increasing the recognition measures. Finally, the system is examined and computed with divergent measures. Hence, the accuracy of the proposed model attains 97.8 for dataset 1, 96.95 for dataset 2, 98.15 for dataset 3 and 96.65 for dataset 4. This highest value proves the system efficiency of differentiating the human emotions. Thus, the enhanced method demonstrates that the system attains extensive results in easily recognizing different emotions.