Universal TTS Model Training & Dataset Preparation Guide

Welcome to a practical guide for preparing speech datasets, training or fine-tuning custom Text-to-Speech models, running inference, and packaging models for reuse.

Start here

Follow the workflow in order if you are starting a new project:

1. Prepare your dataset
2. Set up the training environment
3. Train or fine-tune the model
4. Generate speech with inference
5. Package and share the model
6. Troubleshoot and find resources

Choose your path

• Raw audio, missing transcripts: start with Data Preparation.
• Clean audio-text pairs: continue with Training Setup.
• A trained checkpoint: go to Inference.
• A model ready to distribute: read Packaging and Sharing.

The guide is framework-agnostic where possible. It focuses on the decisions and checks that transfer between modern TTS toolchains.

Use the language selector in the header to switch between the available translations. Contributions and new translations are documented in the Translation Guide.