humair-experiments/tts-pretrain-clones-3m

TTS Pretrain Clones (3M)数据集包含2,967,779个克隆话语，覆盖2,971名英语说话者。数据采样率为44.1 kHz，以WAV格式存储在Parquet文件中。这些数据是通过echo-tts合成英语文本生成的，基于从Qwen3-TTS VoiceDesign基础说话者中提取的说话者潜在特征。每个说话者包含10个语音克隆潜在特征和100个文本。每个说话者的第一个话语（第0行）在伴侣参考数据集中单独发布。覆盖范围包括说话者1-60、61的部分（749行）以及91-3000。由于迁移过程中的存储事故，有三十名说话者（61的尾部以及62-90）缺失，但他们的参考数据仍在伴侣数据集中。伴侣数据集（参考）为：Aynursusuz/tts-pretrain-refs-3k。

TTS Pretrain Clones (3M) dataset contains 2,967,779 clone utterances across 2,971 English speakers. Sample rate: 44.1 kHz, WAV stored in Parquet files. Generated by echo-tts synthesizing English text on speaker latents derived from Qwen3-TTS VoiceDesign base speakers. Per speaker: 10 voice-clone latents × 100 texts. The first utterance of each speaker (row 0) is published separately in the companion refs set. Coverage: speakers 1-60 + 61 (partial, 749 rows) + 91-3000. Thirty speakers (61s tail + 62-90) are absent due to a storage incident during migration; refs for them remain in the companion dataset. Companion dataset (references): Aynursusuz/tts-pretrain-refs-3k.