yilele/synth-qa-taste-codec-chat

Synthetic QA Taste-S Codec Chat 18571 single-turn Traditional Chinese QA utterances with synthesized speech, 21.6 hours of audio before codec extraction. Assistant speech is represented as: <SAY> text_token <a_code> <b_code> ... <p_code> ... </SAY> Each text token is followed by its 16 Taste-S FSQ codes (codebooks a..p). Configurations default — messages (user question + assistant <SAY> speech), audio, and answer text. Statistics Utterances: 18571… See the full description on the dataset page: https://huggingface.co/datasets/yilele/synth-qa-taste-codec-chat.