Home Assistant voice pipeline: Wyoming, Ollama and ESPHome

This overview explains the layers of a local AI voice assistant in Home Assistant. Jarvis, the project's assistant, narrates the video in place of Lars. The focus is on how integrations connect the speaker hardware, speech services and locally hosted language model.

The Wyoming protocol connects transcription and voice synthesis services. Faster Whisper and Piper are presented as the default Home Assistant OS apps. Lars's setup instead uses Onyx ASR with Nvidia Parakeet for transcription and runs Kokoro as a Wyoming service on another computer. The narrator prefers these alternatives, but the video does not provide speech benchmarks.

Home Assistant's Ollama integration configures conversation assistants, model selection and the system prompt. Tools for Assist adds access to search services and weather forecasts. The narrator also describes OpenClaw as an alternative connection through an AI agent harness, claiming more capable conversations at the cost of slower responses.

ESPHome connects smart speakers to Home Assistant and supports on-device wake word detection. The hardware comparison covers Spotpear's single microphone, the ESP32-S3-AUDIO's dual microphone array and FutureProofHomes Satellite1's four microphones and two-way speaker. The narrator reports transcription and sound limitations with Spotpear, while explicitly noting that Satellite1 has not been tested.