Built with Claude Code. See how these projects were built for what that means in practice.

The pipeline that lets viewers trigger AI voice lines on stream:

Viewer types "!tts hello world"
  → Streamer.bot command → C# action
  → POST /tts to the local Chatterbox TTS server
  → Chatterbox generates WAV with a voice-cloned reference
  → Streamer.bot plays it → OBS Desktop Audio picks it up

Voice roster

Tech

• Chatterbox-TTS-Server — MIT-licensed TTS, chatterbox-turbo model (~5GB VRAM, ~78ms time-to-first-audio on the RTX 5090).
• Streamer.bot — each voice is one command + one action + one C# sub-action. The C# template reads args["querystring"] (not rawInput — that one's empty) and posts the text to the TTS server.
• Paralinguistic tags ([laugh], [sigh], …) work on turbo.

Hard-won lessons

• Reference audio must be under 30s or the server rejects it — trim the longer samples.
• Streamer.bot's C# API lacks WebClient.Timeout and CPH.GetVar — compile errors until you work around them.
• flash-attn needs a prebuilt wheel from HuggingFace to work with torch 2.9+cu128.
• Chatterbox beats ElevenLabs in 63.75% of blind preference tests.