starlightmii/gptsovits-retrain-data

🎙️ GPT-SoVITS retrain data — ORIGINAL QUALITY

Source: /run/media/starlight/Data Storage/MOONLIGHT V2/GPT-SoVITS_NEW/datasets/ (Genki-Shadow VO dumps).

Folder	Voice	Files	Format
qiqi/	QiQi (EN)	108 .ogg	Vorbis 48 kHz mono — untouched, 0% loss
hutao_en/	Hu Tao (EN)	188 .ogg	Vorbis 48 kHz mono — untouched, 0% loss

Single archive for one-shot download: gsv_dataset_v2.tar (uncompressed tar, 26.7 MB, sha256 215c8007b1b42ad200870852bf87e423fa4e1997db998c5347c079c52da21fdc).

⚠️ The previous repo (gptsovits_train_data.tar.gz) was lossy-trashed — slices had been resampled to 32 kHz PCM with 36 files missing (qiqi 72/110, hutao 179/237). Models trained on it all sounded noisy. This one is the clean original. Slice in webui 1A, then train.