VOICEVOX/voicevox_engine

What it solves

VOICEVOX ENGINE provides a way to perform text-to-speech (TTS) synthesis via an HTTP server. It allows users and developers to generate high-quality synthetic speech from text, adjust pronunciation and intonation, and integrate these capabilities into other applications without needing to build a TTS engine from scratch.

How it works

The project operates as an HTTP server that exposes a set of APIs for speech synthesis. The process typically involves two steps: first, an /audio_query request is sent to generate a synthesis query (containing parameters like speed and intonation), and second, a /synthesis request is sent using that query to produce the final .wav audio file. It supports both CPU and GPU acceleration (via NVIDIA) and can be deployed using Docker.

Who it’s for

• End users who want to generate synthetic speech for content creation.
• Developers who want to integrate VOICEVOX speech synthesis into their own software via API.
• Engine creators who want to build VOICEVOX-compatible engines to be used within the VOICEVOX editor.

Highlights

• Flexible Control: Adjust speech speed, intonation, and pronunciation using a custom notation (AquesTalk-style).
• User Dictionary: Full API support for adding, editing, and deleting custom words to improve pronunciation accuracy.
• Advanced Synthesis: Includes features like voice morphing (blending two different voices) and singing voice synthesis (using MIDI-based scores).
• Developer Friendly: Provides Docker images for easy setup, a comprehensive API documentation, and support for multiple engine instances (Multi-Engine feature).