TryOnVirtual/tts-pretrain-clones-3m

TTS Pretrain Clones (3M) 是一个用于文本到语音（TTS）预训练的数据集，包含2,967,779个克隆话语，覆盖2,971个英语说话人。音频数据以44.1 kHz采样率的WAV格式存储在Parquet文件中，通过echo-tts合成英语文本生成，基于Qwen3-TTS VoiceDesign基础说话人的潜在变量。每个说话人包含10个声音克隆潜在变量和对应的100个文本。数据集覆盖了说话人1-60、61（部分，749行）以及91-3000，但缺少说话人61的尾部以及62-90（共30个说话人，由于存储迁移事故导致缺失）。配套数据集为Aynursusuz/tts-pretrain-refs-3k，包含每个说话人的第一个话语（第0行）作为参考。

TTS Pretrain Clones (3M) is a dataset for text-to-speech (TTS) pretraining, containing 2,967,779 clone utterances across 2,971 English speakers. The audio data is in WAV format with a 44.1 kHz sampling rate, stored in Parquet files, generated by echo-tts synthesizing English text on speaker latents derived from Qwen3-TTS VoiceDesign base speakers. Each speaker includes 10 voice-clone latents and 100 texts. The dataset covers speakers 1-60, 61 (partial, 749 rows), and 91-3000, but lacks the tail of speaker 61 and speakers 62-90 (30 speakers in total, absent due to a storage incident during migration). The companion dataset, Aynursusuz/tts-pretrain-refs-3k, contains the first utterance (row 0) of each speaker as references.