TwinPeaksTownie/GPT_SoVITS_LAURA_v1

Laura - GPT-SoVITS v2ProPlus Finetuned Voice

Finetuned GPT-SoVITS v2ProPlus weights for the Laura voice.

Files

File	Description
GPT_weights_v2ProPlus/laura-e15.ckpt	S1 (GPT) autoregressive model - handles prosody and timing
SoVITS_weights_v2ProPlus/laura_e8_s120.pth	S2 (SoVITS) VITS model - handles acoustic synthesis
voices/laura.wav	Reference audio for inference
config.json	Training configuration

Usage

1. Install GPT-SoVITS
2. Download the weights and place them in the corresponding directories
3. Load the GPT weights (laura-e15.ckpt) as the S1 model
4. Load the SoVITS weights (laura_e8_s120.pth) as the S2 model
5. Use voices/laura.wav as the reference audio for inference

Training Details

• Base model: GPT-SoVITS v2ProPlus
• GPT checkpoint: Epoch 15
• SoVITS checkpoint: Epoch 8, Step 120
• Batch size: 12
• Learning rate: 0.0001
• Sampling rate: 32kHz
• LoRA rank: 32
• Pretrained S2G: s2Gv2ProPlus.pth
• Pretrained S2D: s2Dv2ProPlus.pth