worstchan/UltraChat-300K-SLAM-Omni

UltraChat-300K数据集是一个用于支持论文SLAM-Omni复现的对话数据集。该数据集包含301,187个训练样本，每个样本包括问题、问题音频、答案、答案的语音合成标记等。数据集经过修改，移除了过长的数据样本，并使用CosyVoice合成了语音响应标记，这些标记作为模型训练的目标。数据集来源于ultrachat。

The UltraChat-300K (Modified) dataset is prepared to support the reproduction of the paper SLAM-Omni: Timbre-Controllable Voice Interaction System with Single-Stage Training. The dataset includes features such as split_name, index, round, question, question_audio, answer, answer_snac, and answer_cosyvoice_speech_token. The dataset has undergone data filtering to remove excessively long samples and synthesized corresponding speech response tokens using CosyVoice. The original version of the dataset is sourced from ultrachat.