Live Demo

Voice Assistant

Continuous voice conversation with Aria — start talking and she responds naturally. Speech recognition and synthesis run locally; the language model follows the configured provider.

English only. Aria understands other languages and answers in them, but every voice available to her is English — so anything else comes back with an English accent. A German voice is on the list.

faster-Whisper STT RAG retrieval Kokoro TTS Browser VAD FastAPI LLM
Ready
Your browser will ask to allow microphone access when you start. Click Allow to enable voice input.
Conversation

Click Start Conversation to talk, or type below.

Aria can answer questions about Timo's projects, demos, and tech stack.

Agent Flow
Browser VAD
Web Audio AnalyserNode detects speech onset via RMS amplitude
↓
Audio Capture
MediaRecorder streams audio on shared mic — 1.5 s silence ends the turn
↓
Speech-to-Text (faster-Whisper)
Audio blob uploaded to /api/speech/stt — returns transcript
↓
RAG + LLM (gpt-oss:20b)
Transcript sent to /api/voice/chat — FAISS retrieval + Aria persona
↓
Text-to-Speech (Kokoro)
Reply synthesised via /api/speech/tts — voice af_heart, 60+ voices available
↓
Audio Playback + Resume
Browser plays MP3 — VAD automatically resumes listening on completion
0 events