GptSoVitsCheckpoints

Models & Types · AiDotNet.TextToSpeech.CodecBased

The files a GPTSoVITS loads (the release’s GPT_SoVITS/pretrained_models, Hugging Face lj1995/GPT-SoVITS).

Constructors

Constructor Summary
GptSoVitsCheckpoints()

Properties

Property Summary
string Bert { get; set; } chinese-roberta-wwm-ext-large’s weights (chinese-roberta-wwm-ext-large/pytorch_model.bin), which Chinese text needs.
string Discriminator { get; set; } The discriminator’s checkpoint for GAN fine-tuning (s2D488k.pth, s2Dv2ProPlus.pth, …).
string Hubert { get; set; } chinese-hubert-base’s weights (chinese-hubert-base/pytorch_model.bin).
string SpeakerVerification { get; set; } ERes2NetV2’s weights (sv/pretrained_eres2netv2w24s4ep4.ckpt), which v2Pro and v2ProPlus need.
string Synthesizer { get; set; } The synthesizer’s checkpoint (an s2 .pth, e.g.
string SynthesizerBase { get; set; } For a LoRA Synthesizer: the pretrained synthesizer it adapts (s2Gv3.pth, s2Gv4.pth, …).
string TextToSemantic { get; set; } The GPT’s checkpoint (an s1 .ckpt, e.g.
string Vocoder { get; set; } The vocoder (v3: models--nvidia--bigvgan_v2_24khz_100band_256x/bigvgan_generator.pt; v4 and V5: gsv-v4-pretrained/vocoder.pth or gsv-v5-pretrained/vocoder.pth).