Push-to-Talk - NeuralCleave

Push-to-Talk (PTT) lets you trigger voice input manually by holding a button in the dashboard. This avoids the always-on microphone access required by wake word mode.

How it works

Hold the PTT button (phone icon) in the chat toolbar or the /voice page.
Speak your message.
Release the button.
NeuralCleave transcribes via STT and generates a reply.
The reply is delivered as text (and played via TTS if configured).

Configuration

PTT requires STT to be configured:

[voice]
stt = "whisper"
stt_model = "base"

No additional PTT config is needed — the feature activates automatically when stt is set.

PTT button states

State	Appearance
Unavailable (STT not configured)	Dimmed, disabled — tooltip: “PTT not configured — enable voice in Settings”
Ready	Normal icon
Recording	Pulsing green background

The PTT button is always visible in the chat toolbar so users can see the feature exists even before configuring voice.

API

You can trigger PTT programmatically:

# Start recording
curl -X POST http://localhost:7432/api/v1/voice/ptt/start
# { "started": true }

# Stop recording and get transcript
curl -X POST http://localhost:7432/api/v1/voice/ptt/stop
# { "transcript": "What is the weather today?", "response": "It is..." }

Compared to wake word

	Wake word	PTT
Hands-free	Yes	No
Always listening	Yes	No (mic only while held)
Privacy	Mic always on	Mic only when pressed
Latency	Higher (handoff + VAD)	Lower (instant)
False triggers	Possible	None