Welcome to my very first attempt at training a custom voice using Piper TTS — an open-source, fast, and high-quality text-to-speech engine.
After 26 hours of training, here's the result! 🎉
🎬 Demo Video:
compressed-prime.mp4
(If the video doesn't load, click here to download or view it directly.)
If you find this plugin useful, please consider donating. Your support is greatly appreciated!
- Engine: Piper TTS
- Training Time: ~26 hours
- Hardware: i7-13700K, 4090, 32GB ram
- Dataset: Optimus Prime Voicelines (Hugging Face)
A huge thanks to srinivasbilla for providing this fantastic dataset of Optimus Prime voice lines.
I’ve always been fascinated by voice synthesis and wanted to try building a voice model from scratch using open tools. This project gave me the chance to learn about:
- Dataset cleaning and formatting
- Training large TTS models with Piper
- Voice evaluation and improvement
- This voice is English-only for now.
- Improve audio quality through better preprocessing
- Adjust pronunciation and timing
- Experiment with additional datasets and multilingual voices
- Share the trained model for others to try
- Piper TTS for the amazing TTS engine.
- ifansnek/piper-train-docker for great working Docker image to train a voice (glad I didnt have to make this)
- Optimus Prime Voicelines Dataset by @srinivasbilla
This project is licensed under the MIT License.