Documentation Index
Fetch the complete documentation index at: /llms.txt
Use this file to discover all available pages before exploring further.
NeuralCleave’s local voice pipeline — wake word, VAD, STT, and TTS, all in-process.
Microphone
↓
OpenWakeWord (wake word detection)
↓ on wake
Voice Activity Detection (VAD — energy or WebRTC)
↓ speech detected
faster-whisper (STT)
↓ transcript
NeuralCleave LLM pipeline (same as text chat)
↓ reply text
TTS (ElevenLabs → Kokoro → pyttsx3)
↓
Speaker output
[voice]
stt = "whisper"
stt_model = "base"
tts_engine = "kokoro"
wake_word = "alexa"
continuous_voice_enabled = true # required alongside wake_word — see Wake Word
/voice shows live status.
| Mode | How to trigger | Use case |
|---|---|---|
| Wake word | Say the wake word | Hands-free, always-on |
| Push-to-talk (PTT) | Hold the PTT button in the UI | Low-latency, no false triggers |
| Continuous listen | Toggle in the Voice page | Voice-first workflows |
| Microphone button | Click mic icon in chat toolbar | One-shot voice input |