Evidence note: Metrics in this note refer to repository tests or controlled scenarios unless a live production source is explicitly linked.
Cloud-based voice assistants (Siri, Alexa) suffer from privacy leakage and 2-second cloud roundtrip latencies. For desktop productivity, developers need a 100% local voice assistant that activates on custom wake words and executes local terminal commands.
I created Opti (optinonymous): a local Python voice assistant combining on-device Porcupine wake word detection, streaming local Whisper STT, Piper neural TTS, and MCP tool execution.
1. Local Processing Pipeline
| Component | Engine | Latency | Privacy |
|---|---|---|---|
| Wake Word | Picovoice Porcupine | < 50 ms | 100% On-Device |
| Speech-to-Text | Whisper.cpp (quantized) | 120 ms | Local CPU/GPU |
| Text-to-Speech | Piper Neural TTS | 80 ms | Local ONNX runtime |
Sources, Code & Further Reading
- • Codebase: Private local voice-assistant implementation; this note documents its architecture without implying a public repository.
- • Piper Neural TTS: github.com/rhasspy/piper — Fast, local neural text-to-speech.