Abstarct: Text-to-Speech (TTS) is a key area in human–machine interaction, aiming to generate natural, fluent, and intelligible speech from textual input. Despite significant advances in TTS technologies, particularly for high-resource languages such as English, Persian remains a challenging language due to its unique phonological, syntactic, and orthographic characteristics. The absence of explicit phonetic markers in written Persian, the prevalence of homographs, flexible word order, and the lack of large-scale standardized speech corpora pose major obstacles to developing high-quality Persian TTS systems. In this thesis, an end-to-end Persian text-to-speech system baxsed on the unified VITS architecture is designed and implemented. This architecture integrates text encoding, alignment learning, acoustic modeling, and waveform generation into a single frxamework, which reduces system complexity and improves the naturalness of synthesized speech. The primary objective of this research is to generate fluent, expressive, and human-like speech while minimizing the need for post-processing steps. To evaluate the performance of the proposed system, the perceptual Mean Opinion Score (MOS) metric is employed. In this evaluation, the synthesized speech samples are compared with natural speech from the reference dataset and assessed by 20 native Persian listeners. The experimental results indicate that the proposed model achieves a satisfactory level of naturalness, fluency, and similarity to human speech. These findings demonstrate the strong potential of the VITS architecture for developing high-quality Persian text-to-speech systems as well as for other low-resource languages with similar linguistic characteristics.