https://blog.landryzetam.net/tags/speech-recognition/Recent content in Speech-Recognition on Landry Zetam - Tech BlogHugo -- gohugo.ioen-usFri, 08 Aug 2025 00:00:00 +0000https://blog.landryzetam.net/posts/voice-pipeline/Fri, 08 Aug 2025 00:00:00 +0000https://blog.landryzetam.net/posts/voice-pipeline/<h2 id="overview">Overview</h2> <p>This implementation demonstrates a complete voice assistant pipeline using Wyoming Satellite with Home Assistant and Ollama as the AI brain, achieving natural conversation and smart home control entirely on-premises. The system transforms spoken words into meaningful actions through a sophisticated pipeline of AI models, all running locally for complete privacy.</p> <h2 id="why-ollama-for-voice">Why Ollama for Voice?</h2> <p>Voice assistants have unique requirements that Ollama handles beautifully:</p> <h3 id="speed-requirements">Speed Requirements</h3> <ul> <li><strong>Response Time</strong>: &lt; 2 seconds for natural conversation</li> <li><strong>Model Loading</strong>: Pre-loaded models for instant responses</li> <li><strong>Streaming</strong>: Real-time token generation for faster perceived response</li> </ul> <h3 id="privacy-first">Privacy First</h3> <ul> <li><strong>Local Processing</strong>: No cloud dependencies</li> <li><strong>Conversation History</strong>: Stored locally, never uploaded</li> <li><strong>Custom Wake Words</strong>: Combined with OpenWakeWord for complete privacy</li> </ul> <h2 id="components-used">Components Used</h2> <ul> <li><strong>OpenWakeWord</strong>: Wake word detection (&ldquo;Hey Jarvis&rdquo;)</li> <li><strong>Whisper</strong>: Speech-to-text transcription</li> <li><strong>Ollama (Qwen2.5:3b &amp; Llama 8B)</strong>: Natural language understanding and response generation</li> <li><strong>Piper TTS</strong>: Text-to-speech synthesis</li> <li><strong>Home Assistant</strong>: Smart home orchestration and pipeline management</li> <li><strong>Wyoming Satellite</strong>: Voice satellite software on Raspberry Pi</li> </ul> <h2 id="hardware-setup">Hardware Setup</h2> <ul> <li><strong>Raspberry Pi 5</strong> (2.4GHz, 4GB RAM) - Voice satellite device</li> <li><strong>ME6S USB Microphone</strong> - Audio input device</li> <li><strong>Bluetooth Speaker</strong> (88:92:CC:5D:CB:79) - Audio output via BlueALSA</li> <li><strong>Mini PC</strong> - Home Assistant server</li> <li><strong>Kubernetes Cluster</strong> - Container orchestration for AI services</li> <li><strong>NVIDIA GPU Node (yeezyai)</strong> - GPU acceleration for Ollama</li> </ul> <h2 id="the-voice-pipeline-architecture">The Voice Pipeline Architecture</h2> <p>The voice pipeline runs entirely on the K8s homelab, with Ollama at its heart:</p>