We have proposed a deep learning-based method for speaker verification. Our team has worked on this project for over a year and the accuracy exceeded criteria such as the accuracy of the Andrew Zisserman group paper from the university of oxford. Unlike other text-dependent methods, our system is text- and language-independent. Our model's processing speed ​​is less than 1 sec. The system identifies a person by simply checking two voices lasting 4 seconds. Our system works perfectly in English, French, Spanish, German, Persian and Arabic, even in a noisy environment.