This dataset contains structured Hungarian conversations generated with multiple large language model families for the study “Efficient ASR Training with Conversations that Never Happened.” Paper: arXiv link
Each model is provided as a separate Parquet file. The dataset contains the generated textual conversations and associated scenario and participant metadata; it does not contain synthesized audio.
