Discussion of a master's thesis in the College of Computer Science and Mathematics - Department
Discussion of a master's thesis in the College of Computer Science and Mathematics - Department of Computer Science
Entitled (Detecting the speaker's language in an audio file using intelligent techniques)
On Sunday (8-5-2022), the College of Computer Science and Mathematics at the University of Mosul was discussed in a master's thesis tagged (Detecting the speaker's language in an audio file using intelligent techniques).
For the student: Muhammad Nayef Abd Abdullah and under the supervision of Prof. Dr. Fawzia Mahmoud Ramo
The thesis submitted by the student dealt with a proposal for a system called (Detecting Speaker's Language using Deep Learning (DSDL)) in order to detect the speaker's language. It was relied on a database being the researcher named (M2L-Dataset) consisting of four languages (Arabic, English, French and Kurdish). ).
The study dealt with the use of deep learning algorithms, where three models of the system were proposed, the first model is the convolutional neural network algorithm to deal with images of spectral diagrams, the second model is the convolutional neural network algorithm to deal with the dictionary file, and the third model is the bidirectional long-short-term memory network algorithm to deal with the dictionary file.
The study aims to build a local database for a number of people of different ages, for both sexes, and for the two official languages in Iraq (Arabic and Kurdish), in addition to my languages (English and French), and support the global database and upload it to the global website (Kaggle) to make data available to all researchers. Using the best methods of character extraction and initialization of the audio file using the Tone Coefficient Algorithm (MFCC). As well as the use of more than one method to represent the properties extracted from the audio file, where the representation of the dictionary file (JSON) and images of spectrograms were used, which led to reaching the best results. Using more than one intelligent algorithm to detect the type of language used in the message to show the efficiency of each method and the suitability of each method to the type of data.
The discussion committee was chaired by Assistant Professor Dr. Muhammad Gajan Younes and the membership of Assistant Professor Aseel Walid Ali and Professor Dr. Mayada Ramez Mahmoud from Dohuk Governorate / Zakho University and under the supervision and membership of Prof. Dr. Fawzia Mahmoud Ramo