GptSoVitsVersion

Enums · AiDotNet.TextToSpeech.CodecBased

The GPT-SoVITS release a GPTSoVITS follows (each has its own released weights).

Fields

Field Summary
GptSoVitsVersion V1 v1: Chinese, Japanese and English with v1’s 322 symbols and pypinyin; the GAN synthesizer.
GptSoVitsVersion V2 v2: the 732-symbol set adding Korean and Cantonese, G2PW for Chinese; the GAN synthesizer.
GptSoVitsVersion V2Pro v2Pro: v2 with an ERes2NetV2 speaker embedding added to the style vector.
GptSoVitsVersion V2ProPlus v2ProPlus: v2Pro with a wider generator (the default).
GptSoVitsVersion V3 v3: the text encoder’s hidden state to 24 kHz mels by conditional flow matching, then BigVGAN.
GptSoVitsVersion V4 v4: flow matching to 32 kHz-hop mels, then a 48 kHz HiFi-GAN vocoder.
GptSoVitsVersion V5Dev V5 dev: v4’s design without the step embedding, with classifier-free guidance and rolling-tail chunks.
GptSoVitsVersion V5Turbo V5 turbo: V5 sampled in four steps without guidance.
int value__