In this paper, we explore how large language models (LLMs) can be integrated into translation workflows through multimodal interaction. Recent developments in generative Artificial Intelligence (AI) are influencing translation workflows, yet current uses of LLMs in translation and post-editing (PE) remain mostly outputoriented, providing limited opportunities for users to guide the editing process. We present COPECO-Speech1, a multimodal teaching platform for translation and PE. It allows users to interact with the translation process through typing, speech-based navigation commands and spoken natural-language instructions interpreted by an LLM. The system is web-based and combines automatic speech recognition (ASR), machine translation (MT), and LLM-based editing within a single workflow, allowing more interactive translation and PE. COPECO-Speech was designed for both research and teaching purposes, and supports the collection of interaction data such as speech logs, keystrokes, timing information, and PE.