Summary
VOICEROID2 driven from the conversation.
The kana step is exposed on its own, so when the engine reads a name wrong you fix the reading and speak it again rather than fighting the text.
What it is
A Node.js MCP server in front of voiceroid_daemon, which drives a local VOICEROID2 installation. Text goes in, audio comes out and plays on macOS, Windows or Linux; the kana reading can be generated separately and corrected before it is spoken.
What you get
- Japanese text spoken in a VOICEROID2 voice and played back
- Text converted to its kana reading, which you can correct and pass back in
- Volume, speed, pitch, emphasis and three pause lengths set per call or as defaults
- A connection check against the daemon before anything else is attempted
Requirements
Node.js 18 or higher, VOICEROID2 installed, and voiceroid_daemon running — the server looks for it on localhost by default, and its address is configurable. Basic authentication is supported where the daemon asks for it.
Setup effort
Build from source — clone the repository and build it, then point your client at the binary