FAQ & troubleshooting

Does whispa capture the microphone or play audio?

No — whispa orchestrates the STT → LLM → TTS core. Capturing mic audio and playing the WAV are out of scope (use sox/arecord/ffmpeg, or see whispa-pi for a ready appliance).

Can I develop without whisper.cpp / Ollama / Piper installed?

Yes. Use the built-in MockStt, EchoLlm, MockTts for tests and demos; swap in the real adapters when you're ready.

WhisperCppStt fails / no output

Ensure whisper-cli is on your PATH and the model path is correct, and that the input is a 16 kHz mono WAV. Pass a custom binary if your build names it differently.

OllamaLlm can't connect

Make sure ollama serve is running and the model is pulled (ollama pull llama3.2). Default host is http://localhost:11434; override with the host option.

How do I add a different engine (cloud STT, another LLM)?

Implement the SttEngine / LlmEngine / TtsEngine interface and pass it to VoicePipeline. The loop and history stay the same. See architecture.

How is conversation history managed?

VoicePipeline keeps a Message[] (system prompt first). maxHistory bounds it; reset() clears everything except the system prompt.