zuhri025/high-quality-unified

Audio Token Unified 是一个统一合并的数据集，专门用于文本转语音（TTS）训练。该数据集由三个来源合并而成：humair025/Magpie-Speech-Orpheus-125k-annotated、humair025/DeepDialogue-orpheus-Kanade-Annotated 和 humair025/DeepDialogue-xtts-K-Annotated。数据集包含以下列：audio_duration（音频时长）、text（文本）、audio_global_embedding（音频全局嵌入）、audio_content_token_indices（音频内容令牌索引）和audio_token_len（音频令牌长度）。格式为单个parquet文件，针对TTS训练进行了优化，并减少了不必要的元数据，以提高训练效率和性能。数据集支持英语和乌尔都语，适用于多语言文本转语音任务。Audio Token Unified is a unified merged dataset specifically designed for text-to-speech (TTS) training. It is created by merging three sources: humair025/Magpie-Speech-Orpheus-125k-annotated, humair025/DeepDialogue-orpheus-Kanade-Annotated, and humair025/DeepDialogue-xtts-K-Annotated. The dataset includes the following columns: audio_duration, text, audio_global_embedding, audio_content_token_indices, and audio_token_len. It is formatted as a single parquet file, optimized for TTS training, and reduces unnecessary metadata to enhance training efficiency and performance. The dataset supports English and Urdu languages, making it suitable for multilingual text-to-speech tasks.