Work

การพัฒนาโมเดลการเลียนเสียงเชิงลึกในการประยุกต์ใช้งานด้านสงครามไซเบอร์

Year: 2023
Type: article
Abstract: The objectives of this research were 1) to study the implementations of a deep voice technology, 2) to investigate an appropriate model for the development of a deep voice model for Thai language use, 3) to analyze and evaluate the efficiency of the deep voice model after being used by the Internet users, and 4) to present guidelines for using the deep voice technology in the field of cyber warfare. This study showed that the model for text-to-speech synthesis could be employed in the application of a deep voice technology (Voice Cloning using Deep Learning) in general work, especially in the case of English speech. However, the model had significant limitations including the requirements of the parallel data and the preparation time required for a Thai-language model training. In this work, the researcher had studied and presented guidelines to develop and apply deep voice models for cyber warfare using a GAN (Generative Adversarial Networks) model to overcome the limitations of human speech synthesis models from text sentences. The results revealed that the StarGAN-VC (StarGAN Voice Conversion) and CycleGAN-VC (CycleGAN Voice Conversion) models could be used to transform the voices of ordinary people into those of the targeted people such as politicians and national executives to create fake news in cyber warfare. Moreover, spoofed voices generated from this model could achieve a highest mean opinion score (MOS) of 3.59 and had the potential to deceive Internet users into believing that the spoofed voice was the voice of a real target. In most severe cases, 40% of the samples were duped by fake voices. These findings highlight and raise awareness of new threats that might affect national security in the future, and the search for ways to detect and prevent them.
Language: Thai
Open Access status: diamond
OpenAlex ID: W7219546464
FWCI:
Cites: 13
Cited by:
Related to: 10