A network architecture for building applications that use speech recognition and/or synthesis

International audienceThis paper proposes a simple versatile architecture for sharing speech recognition and synthesis resources in a heterogeneous networked environment. Using a unified application programming interface and the concept of proxies allows many clients to gain access to services that would normally be reserved to specific computers. This helps reducing the total cost of ownership while providing sophisticated service through simple approaches. At the same time, multi-speaker data collecting will be simplified. We show that using this architecture for speech activated software reduces the requirements of client computers while adding a negligible latency compared to a powerful workstation