DailyTalkContiguous-ja is a synthetic multi-turn Japanese conversational speech dataset in which DailyTalk [Keon Lee etal., 2022]
translated by Gemma-3-27B and speech data is synthesized by TTS engine Zyphra/Zonos-v0.1-transformer.
For each speaker in covnersation, different voice is randomly asssigned from voice dataset with five voices in total.
As like with kyutai/DailyTalkContiguous, rather than having separate files… See the full description on the dataset page:
https://huggingface.co/datasets/Rakuto/DailyTalkContiguous-ja.
