Search

Speech emotion recognition (SER) has gained an increased interest during the last decades as part of enriched affective computing. As a consequence, a variety of engineering approaches have been developed addressing the challenge of the SER problem, exploiting different features, learning algorithms, and datasets. In this paper, we propose the application of the graph theory for classifying emotionally-colored speech signals. Graph theory provides tools for extracting statistical as well as structural information from any time series. We propose to use the mentioned information as a novel feature set. Furthermore, we suggest setting a unique feature-based identity for each emotion belonging to each speaker. The emotion classification is performed by a Random Forest classifier in a Leave-One-Speaker-Out Cross Validation (LOSO-CV) scheme. The proposed method is compared with two state-of-the-art approaches involving well known hand-crafted features as well as deep learning architectures operating on mel-spectrograms. Experimental results on three datasets, EMODB (German, acted) and AESDD (Greek, acted), and DEMoS (Italian, in-the-wild), reveal that our proposed method outperforms the comparative methods in these datasets. Specifically, we observe an average UAR increase of almost 18 % , 8 % and 13 % , respectively. [ABSTRACT FROM AUTHOR]

Speech emotion recognition (SER) has recently increased because of vast innovations in human–computer interaction and affective computing. In recent years, numerous deep learning-based schemes presented for SER have shown significant improvement over the traditional machine learning approaches. Most deep learning-based faced SER systems face challenges due to data imbalance problem that occurs due to unequal samples in the database. The input to two-dimensional CNN uses traditional MFCC for SER. It degrades the quality of deep attributes because of the higher variance, frequency resolution problem and spectral leakage problem of traditional MFCC. This paper proposed the novel Multi-taper Mel Frequency Logarithmic Spectrogram to enrich the Deep Convolutional Neural Network effectiveness for SER. Further, Generative Adversarial Network is used for speech emotion data augmentation during training to deal with data scarcity problems in SER. The performance of the proposed SER scheme is validated using the Berlin EmoDB and RAVDESS datasets. The proposed method provides SER accuracy of 96.65% and 97.12% for the EmoDB and RAVDESS dataset, respectively, and significantly improves over the recent techniques. [ABSTRACT FROM AUTHOR]

Speech emotion recognition (SER) is a vital component of the human–computer interaction system. The traditional Deep learning-based speech SER schemes show poor time-domain representation, class imbalance issues due to uneven samples in the training datasets, less feature distinctiveness, and inferior long-term dependency on global and local attributes of the speech. This article introduces lightweight, long short memory (LSTM) along with multiple acoustic features such as Mel frequency spectrum coefficients (MFCC), chroma, root mean square (RMS), and Tonnetz features for boosting the time-domain representation and long-term dependency of emotional speech. Further, it used data augmentation techniques such as noise addition, shifting, stretching, and feature extraction to lessen the class imbalance problem. The proposed methods have an average accuracy of 95% for SAVEE, 94.57% for EMOVO, 97.16% for EMODB, and 96.66% for BAVED, which has shown noteworthy improvement over the traditional state of arts. [ABSTRACT FROM AUTHOR]

Humans' ability to manage their emotions has a big impact on their ability to plan and make decisions. In order to better understand people and improve human–machine interaction, researchers in affective computing and artificial intelligence are investigating the detection and recognition of emotions. However, different cultures have distinct ways of expressing emotions, and the existing emotion recognition datasets and models may not effectively capture the nuances of the Indian population. To address this gap, this study proposes custom-built lightweight Convolutional Neural Network (CNN) models that are optimized for accuracy and computational efficiency. These models are trained and evaluated on two Indian emotion datasets: The Indian Spontaneous Expression Dataset (ISED) and the Indian Semi Acted Facial Expression Database (iSAFE). The proposed CNN model with manual feature extraction provides remarkable accuracy improvement of 11.14% for ISED and 4.72% for iSAFE datasets as compared to baseline, while reducing the training time. The proposed model also surpasses the accuracy produced by pre-trained ResNet-50 model by 0.27% ISED and by 0.24% for the iSAFE dataset with significant improvement in training time of approximately 320 s for ISED and 60 s for iSAFE dataset. The suggested lightweight CNN model with manual feature extraction offers the advantage of being computationally efficient and more accurate compared to pre-trained model making it a more practical and efficient solution for emotion recognition among Indians. [ABSTRACT FROM AUTHOR]