Emotion Recognition from Spontaneous Tunisian Dialect Speech

Indonesian Journal of Electrical Engineering and Informatics (IJEEI), 2024

Emotions in speech are considered a basic principle of human interaction and play an important role in decision-making, learning, and everyday communication. Research on speech emotion recognition is still being carried out by many researchers to develop speech emotion recognition models with better performance. In this research, we combine the application of data augmentation techniques (Add Noise, Time Stretch, and Pitch Shift) to increase the data size of the Javanese Speech Emotion Database (Java-SED). Mel Frequency Cepstral Coefficients (MFCC) is used for feature extraction and then builds a Convolutional Neural Network (CNN) model and applies Multilayer Perceptron (MLP) to classify human emotions from sound. In this research, we produced eight experimental models with a combination of different augmentation techniques. The CNN model parameters include 40 input neurons, four hidden layers with varying neuron counts, Relu activation functions, L2 regularization, dropout rates, Adam optimization, and ModelCheckpoint callbacks to save the best model based on validation loss. From the results of the evaluation that has been carried out, the CNN algorithm produces the highest performance with an accuracy of 96.43%, recall of 96.43%, precision of 96.57%, F1-score of 96.48%, and kappa of 95.71% by applying the Add Noise technique, Time Stretch, and Pitch Shift.

Radioelectronic and Computer Systems

Subject matter: Speech emotion recognition (SER) is an ongoing interesting research topic. Its purpose is to establish interactions between humans and computers through speech and emotion. To recognize speech emotions, five deep learning models: Convolution Neural Network, Long-Short Term Memory, Artificial Neural Network, Multi-Layer Perceptron, Merged CNN, and LSTM Network (CNN-LSTM) are used in this paper. The Toronto Emotional Speech Set (TESS), Surrey Audio-Visual Expressed Emotion (SAVEE) and Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) datasets were used for this system. They were trained by merging 3 ways TESS+SAVEE, TESS+RAVDESS, and TESS+SAVEE+RAVDESS. These datasets are numerous audios spoken by both male and female speakers of the English language. This paper classifies seven emotions (sadness, happiness, anger, fear, disgust, neutral, and surprise) that is a challenge to identify seven emotions for both male and female data. Whereas most have wor...

Speech emotions recognition has become one of the active research areas in speech processing and human computer interaction based applications. An experimental study is conducted in this paper with two spectral features (Rasta-PLP and MFCC), four different emotions (Angry, Happy, Neutral and Sad) and five different learning classifiers (K-star, Logit boost, J48, LAD Tree and Random forest). The focus of this work is to evaluate the performance of learning classifiers in term of classification accuracy and analyzing the effect of emotion and spectral features (MFCC and Rasta PLP) on spoken utterances in Urdu language with four different emotions. Demonstrative experiments were conducted in to two phases. Experimental results of the first phase shows that the combination of spectral features provides significant classification accuracy of learning classifiers as compared to the both features are considered individually. Whereas, Two-way ANOVA testing in the second phase of experiments show that both the factors, emotions and features as well as their interaction have significant effect on speech as the P value is less than the confidence interval.

IRJET, 2021

The human voice can be characterized by several attributes such as pitch, timbre, loudness, and vocal tone. It has often been observed that humans express their emotions by varying different vocal attributes during speech generation. This paper presents an algorithmic approach for detection of human emotions with the help speech .The prime objective of this paper is to recognize emotions in speech and classify them in 6 emotion output classes namely angry, fear, disgust, happy, sad and neutral. The proposed approach is based upon the Mel Frequency Cepstral coefficients (MFCC) uses Crema-D database of emotional speech. Data Augmention is perfomed on input data audio file, such as Noise, High Speed, Low Speed etc. are added, thus more the varied data is available to the model better the model understands. Feature extraction is done using MFCC and then the extracted features are Normalized(for Independent Variable), Label Encoding(for Dependent Variable(for SVM,RF)),One Hot Encoding(for Dependent Variable(for CNN)) is done. After this the dataset is divided into Train, Test and given to different models such as Convolutional Neural Network(CNN),Support Vector Machine(SVM), Random Forest(RF) for Emotion prediction. We report accuracy, f-score, precision and recall for the different experiment settings we evaluated our models in. Convolutional Neural Network(CNN) was found to have the highest accuracy and predicted correct emotion 88.21% of the time. Hence, deduction of human emotions through speech analysis has a practical plausibility and could potentially be beneficial for improving human conversational and persuasion skills.

Emotion is considered to be one of the significant human interaction factors. Recognizing human behavior based on emotion is often misinterpreted. However, that has not prevented the analysts from attempting to extract the information from a speech called speech recognition. In this paper, we analyzed the different algorithm's performance in the field of speech signal classification. Considering accuracy and precision as factors, ANN showed higher results with around 80% accuracy and correctness, followed by SVM and CNN competing for each other for accuracy in between 75 and 80 percent. SVM showed close approximation between accuracy and precision, leaving random forest classifier in the last place. Our experimentation showed neural networks performed substantially in the field of speech and signal processing.

International Journal for Research in Applied Science and Engineering Technology Ijraset, 2020

This paper is an effort at developing a Speech Emotion Identifier model by implementing Librosa and sklearn libraries on a RAVDESS dataset. It gives the reader an insight on the way to detect a human’s emotion based on the speech, taken as input audio file. A newly developed speech signal model is applied to provide the user with the likelihood that the given speech is a response to a given emotion. This particular model is built using convolution neural networks (CNN) and classifiers namely Decission Tree, Random Forest and Multi-Layer Perceptron This model finds its applications in various real-world scenarios and therefore the most potent example for the identical would be in Customer care services where the staffs keep changing their way of pitching by recognizing Customers’ emotion from their speech so as to improve their quality of services provided. This paper presents the feasibility of extraction of MFCC features within the model. This model takes into consideration three different classifiers MLP, Random Forest and Decision Tree and by taking a combination of these three, we get the best possible accuracy as output.

Advances in Robotics, Automation and Data Analytics

The research community's ever-increasing interest in studying humancomputer interactions (HCI), systems deducing, and identifying a speech signal's emotional aspects has emerged as a hot research topic. Speech Emotion Recognition (SER) has brought the development of automated and intelligent analysis of human utterances to reality. Typically, an SER system focuses on extracting the features from speech signals such as pitch frequency, formant features, energyrelated and spectral features, tailing it with a classification quest to understand the underlying emotion. The key issues pivotal for a successful SER system are driven by the proper selection of proper emotional feature extraction techniques. In this paper, Mel-frequency Cepstral Coefficient (MFCC) and Teager Energy Operator (TEO) along with a new proposed Feature Fusion of MFCC and TEO referred to as Teager-MFCC (TMFCC) is examined over a multilingual database consisting of English, German and Hindi languages. Deep Neural Networks have been used to classify the different emotions considered, happy, sad, angry, and neutral. Evaluation results show that the proposed fusion TMFCC with a recognition rate of 92.7% outperforms TEO and MFCC. With TEO and MFCC configurations, the recognition rate has been found as 88.5% and 90.0%, respectively.

Sensors

Vocal emotion recognition (VER) in natural speech, often referred to as speech emotion recognition (SER), remains challenging for both humans and computers. Applied fields including clinical diagnosis and intervention, social interaction research or Human Computer Interaction (HCI) increasingly benefit from efficient VER algorithms. Several feature sets were used with machine-learning (ML) algorithms for discrete emotion classification. However, there is no consensus for which low-level-descriptors and classifiers are optimal. Therefore, we aimed to compare the performance of machine-learning algorithms with several different feature sets. Concretely, seven ML algorithms were compared on the Berlin Database of Emotional Speech: Multilayer Perceptron Neural Network (MLP), J48 Decision Tree (DT), Support Vector Machine with Sequential Minimal Optimization (SMO), Random Forest (RF), k-Nearest Neighbor (KNN), Simple Logistic Regression (LOG) and Multinomial Logistic Regression (MLR) wit...

This paper presents a machine learning approach to automatic recognition of human emotions from speech. The approach consists of three steps. First, numerical features are extracted from the sound database by using audio feature extractor. Then, feature selection method is used to select the most relevant features. Finally, a machine learning model is trained to recognize seven universal emotions: anger, fear, sadness, happiness, boredom, disgust and neutral. A thorough ML experimental analysis is performed for each step. The results showed that 300 (out of 1582) features, as ranked by the gain ratio, are sufficient for achieving 86% accuracy when evaluated with 10 fold cross-validation. SVM achieved the highest accuracy when compared to KNN and Naive Bayes. We additionally compared the accuracy of the standard SVM (with default parameters) and the one enhanced by Auto-WEKA (optimized algorithm parameters) using the leave-one-speaker-out technique. The results showed that the SVM enhanced with Auto-WEKA achieved significantly better accuracy than the standard SVM, i.e., 73% and 77% respectively. Finally, the results achieved with the 10 fold cross-validation are comparable and similar to the ones achieved by a human, i.e., 86% accuracy in both cases. Even more, low energy emotions (boredom, sadness and disgust) are better recognized by our machine learning approach compared to the human. Povzetek: Prepoznavanje čustev iz govora s pomočjo strojnega učeanja

Proceedings of the 10th International Conference on Agents and Artificial Intelligence, 2018

In this paper we compare different approaches for emotions recognition task and we propose an efficient solution based on combination of these approaches. Recurrent neural network (RNN) classifier is used to classify seven emotions found in the Berlin and Spanish databases. Its performances are compared to Multivariate linear regression (MLR) and Support vector machine (SVM) classifiers. The explored features included: melfrequency cepstrum coefficients (MFCC) and modulation spectral features (MSFs). Finally results for different combinations of the features and on different databases are compared and explained. The overall experimental results reveal that the feature combination of MFCC and MS has the highest accuracy rate on both Spanish emotional database using RNN classifier 90,05% and Berlin emotional database using MLR 82,41%.