Interaction with robots often requires a speech interface. Particularly robots with many concurrent processes, need a resource efficient solution. The available frameworks are often tightly coupled, making them less suitable for an deep integration. This work presents a modular prototype speech interface based on the Robot Operating System. The individual modules are interchangeable and logically separated. Special attention is paid to the weak hardware of a Raspberry Pi 4. Benchmarks are used to show that this platform is suitable for a real-time speech interface. This does not apply to all software components. Especially GPU-optimized components reach their limits and miss real-time.