Tag
This paper investigates using text-to-speech (TTS) to generate synthetic training data for spoken question answering in Luxembourgish, a low-resource language, and evaluates multi-source TTS configurations with a parameter-efficient SLAM-style architecture.