Development of a Multilingual Vocalization Methodology for Algerian SMS

We have developed an innovative methodology for the multilingual vocalization of Algerian SMS messages, drawing insights from an initial study that highlights the challenges posed by the complex and specific language used in these messages. In the course of our research, we worked with a dataset comprising 20,000 SMS messages, which, after segmentation, yielded 27,903 French words (class fr) and 42,300 Algerian words (class DZ). This dataset served as the foundation for our comprehensive workflow, outlining the various steps involved in implementing a vocalization system tailored to Algerian SMS messages. Our approach begins with data preprocessing, where we focus on text normalization and the extraction of textual features from these messages. Subsequently, the second phase of our methodology centers on developing a language identification system, recognizing that Algerian SMS messages often encompass multiple languages, including French and Arabic. To tackle this challenge, we employed a range of classifiers, including SVM, KNN, MNB, RFC, RNN, and CNN. Notably, SVM achieved the highest accuracy of 62.56% without SMS normalization. However, it's worth highlighting that the accuracy significantly improved to 73.18% after applying text normalization when using the CNN method. Following the language identification step, we proceed to perform speech synthesis based on the textual content, followed by unit concatenation to generate vocalized SMS messages. The results obtained through our methodology have provided valuable insights into both the strengths and limitations of our developed system, setting the stage for promising avenues of future research in the field of multilingual vocalization of SMS messages in dialects.