The purpose of this study is to explore the application and evaluation of multimodal translation technology in French speech translation system. Through the use of multimodal, variational neural machine translation, contrast learning, and deep learning methods, this study proposes an innovative French speech translation system to improve the accuracy and fluency of translation. Firstly, this study adopts a multimodal approach, which fuses multiple modal information such as speech, text, and images together in order to understand the content and context of the source language more comprehensively. Secondly, this study uses variational neural machine translation method to realize automatic translation from source language to target language by constructing an end-to-end neural network model. This study also uses the contrast learning method to find the optimal translation result by comparing the output of different translation models. With real-time speech recognition, intelligent translation, and a visual interface, the system provides a powerful tool for French speech translation, improving the accuracy and efficiency of translation. The system can also learn and optimize based on user feedback, continuously improving the quality and adaptability of translation. The application and effect evaluation of multimodal translation technology in French speech translation system show that it provides an effective solution for French speech translation.