In our daily life, every one of us face different situations and the outcome of it is developing a feeling about it. Emotion is a strong feeling about human's situation or relation with others [1]. It has a big role in customer decision in many domains including e-commerce, restaurants, movies, interests, and satisfaction with a service or a product. Moreover, it affects our health! Lately, Facebook added some reactions including angry, happiness, love, and surprise to allow users to express their emotions toward a comment, picture or an event. In research, emotional analysis is regarded as a sort of higher, evolved form of sentiment analysis. Sentiment analysis aims to classify texts (posts, sentences, or documents) into negative, positive or neutral. Emotional analysis, on the other hand, is a more elaborated, deeper analysis of users' emotions that tries to inspect the psychology of different Abstract Nowadays, sharing moments on social networks have become something widespread. Sharing ideas, thoughts, and good memories to express our emotions through text without using a lot of words. Twitter, for instance, is a rich source of data that is a target for organizations for which they can use to analyze people's opinions, sentiments and emotions. Emotion analysis normally gives a more profound overview of the feelings of an author. In Arabic Social Media analysis, nearly all projects have focused on analyzing the expressions as positive, negative or neutral. In this paper we intend to categorize the expressions on the basis of emotions, namely happiness, anger, fear, and sadness. Different approaches have been carried out in the area of automatic textual emotion recognition in the case of other languages, but only a limited number were based on deep learning. Thus, we present our approach used to classify emotions in Arabic tweets. Our model implements a deep Convolutional Neural Networks (CNN) trained on top of trained word vectors specifically on our dataset for sentence classification tasks. We compared the results of this approach with three other machine learning algorithms which are SVM, NB and MLP. The architecture of our deep learning approach is an end-to-end network with word, sentence, and document vectorization steps. The deep learning proposed approach was evaluated on the Arabic tweets dataset provided by SemiEval for the EI-oc task, and the results-compared to the traditional machine learning approaches-were excellent.

Several high-resource Text to Speech (TTS) systems currently produce natural, well-established human-like speech. In contrast, low-resource languages, including Arabic, have very limited TTS systems due to the lack of resources. We propose a fully unsupervised method for building TTS, including automatic data selection and pre-training/fine-tuning strategies for TTS training, using broadcast news as a case study. We show how careful selection of data, yet smaller amounts, can improve the efficiency of TTS system in generating more natural speech than a system trained on a bigger dataset. We adopt to propose different approaches for the: 1) data: we applied automatic annotations using DNSMOS, automatic vowelization, and automatic speech recognition (ASR) for fixing transcriptions' errors; 2) model: we used transfer learning from high-resource language in TTS model and fine-tuned it with one hour broadcast recording then we used this model to guide a FastSpeech2-based Conformer model for duration. Our objective evaluation shows 3.9% character error rate (CER), while the ground truth has 1.3% CER. As for the subjective evaluation, where 1 is bad and 5 is excellent, our FastSpeech2-based Conformer model achieved a mean opinion score (MOS) of 4.4 for intelligibility and 4.2 for naturalness, where many annotators recognized the voice of the broadcaster, which proves the effectiveness of our proposed unsupervised method.

No abstract