Shiori_reviewed_ft_StyleTTS2

StyleTTS2 Fine-tuned Model

This model is a fine-tuned version of StyleTTS2, containing all necessary components for inference.

Model Details

• Base Model: StyleTTS2-LibriTTS
• Architecture: StyleTTS2
• Task: Text-to-Speech
• Last Checkpoint: epoch_2nd_00004.pth

Training Details

• Total Epochs: 5
• Completed Epochs: 4
• Total Iterations: 1225
• Batch Size: 2
• Max Length: 620
• Learning Rate: 0.0001
• Final Validation Loss: 0.400027

Model Components

The repository includes all necessary components for inference:

Main Model Components:

• bert.pth
• bert_encoder.pth
• predictor.pth
• decoder.pth
• text_encoder.pth
• predictor_encoder.pth
• style_encoder.pth
• diffusion.pth
• text_aligner.pth
• pitch_extractor.pth
• mpd.pth
• msd.pth
• wd.pth

Utility Components:

• ASR (Automatic Speech Recognition)
  • epoch_00080.pth
  • config.yml
  • models.py
  • layers.py
• JDC (F0 Prediction)
  • bst.t7
  • model.py
• PLBERT
  • step_1000000.t7
  • config.yml
  • util.py

Additional Files:

• text_utils.py: Text preprocessing utilities
• models.py: Model architecture definitions
• utils.py: Utility functions
• config.yml: Model configuration
• config.json: Detailed configuration and training metrics

Training Metrics

Training metrics visualization is available in training_metrics.png

Directory Structure

├── Utils/ │ ├── ASR/ │ ├── JDC/ │ └── PLBERT/ ├── model_components/ └── configs/

Usage Instructions

1. Load the model using the provided config.yml
2. Ensure all utility components (ASR, JDC, PLBERT) are in their respective directories
3. Use text_utils.py for text preprocessing
4. Follow the inference example in the StyleTTS2 documentation