gavinlaw/gigaspeech_speaker_prompts_v2

GigaSpeech Speaker Prompts v2 数据集是一个包含 9,989 个单段英语语音提示的集合，这些提示是从 GigaSpeech 语料库中精选出来的，用于零样本语音克隆参考（如 CosyVoice3、XTTS 等）。与之前的 v1 版本相比，v2 版本强制每个独特的 GigaSpeech opus 只能提供一个片段，从而显著提高了声学多样性。数据集包含一个压缩包，内有音频文件和元数据，适用于零样本 TTS 流程。数据集采用 Apache-2.0 许可证，研究使用时需引用原始 GigaSpeech 论文。

The GigaSpeech Speaker Prompts v2 dataset is a pool of 9,989 single-segment English voice prompts curated from the GigaSpeech corpus for use as zero-shot voice-cloning references (e.g., CosyVoice3, XTTS). Compared to the v1 pool, this v2 version enforces one segment per unique GigaSpeech opus, dramatically increasing acoustic diversity. The dataset ships as a tarball containing audio files and metadata, intended for zero-shot TTS pipelines. It is licensed under Apache-2.0 and requires citation of the original GigaSpeech paper for research use.