Requirement formalization represents an indispensable component of software analysis. However, due to the specialized expertise required to formulate the requirements, and the prevalent preference for using natural language to describe them, domain experts often need to iteratively refine and extract formal specifications from unstructured natural language descriptions. This process is inherently challenged by the complexity and inherent ambiguity of natural language, making the construction of accurate and complete formal specifications a non-trivial task. To address this issue, this paper proposes TempVeri, a novel and end-to-end framework focusing on translating natural language requirements to formal specifications in temporal logics. One of the core innovations of TempVeri lies in integrating structured prompt engineering with automated function calling to enable high-quality translation of formal specifications. Furthermore, in order to ensure the correctness of the translation, an ``LLM-as-a-Judge'' semantic validation mechanism is further introduced to assess the semantic consistency of the generated outputs. The proposed framework exhibits strong extensibility and portability. We have also integrated the proposed TempVeri framework into a runtime verification tool and evaluated its correctness and effectiveness using a realistic industrial case study. The experimental evaluations demonstrate the superiority of TempVeri in requirement formalization, confirming its practical utility and potential for broader application in software analysis.