Real-time SV2TTS: Transfer Learning for Multispeaker Text-to-Speech

This open-source project implements real-time multispeaker text-to-speech (SV2TTS) synthesis using transfer learning from speaker verification, based on the author's master's thesis. It's a three-stage deep learning framework: creating a digital voice representation from short audio clips, then using this representation to generate speech from arbitrary text. While the project is older and may have lower quality than commercial alternatives, it supports Windows and Linux, with GPU acceleration recommended. Detailed installation and usage instructions are provided, along with support for various datasets.