Verbi

Why we included this project

Verbi is less a ready-to-ship voice assistant than a workbench for comparing voice models. Each step of the pipeline, transcription, response generation, and text-to-speech, is a separate swappable module, so you can run the same request through OpenAI, Groq, ElevenLabs, Cartesia, or Deepgram, or drop in local options like Ollama, MeloTTS, and Piper when you want to work offline. That setup makes it useful for testing which transcription or voice model performs best on a particular language or accent before you commit to one. Mic input and audio playback are built in, and one config file controls the whole pipeline, so you can have a working voice loop running quickly and then swap models to compare them side by side.