Voice & Telephony

Real-time voice input and output. Four channels cover the matrix: inbound calls, outbound speech synthesis, local microphone wake, and SIP-grade real-time conversation.

ClawdTalk (real-time SIP)

[channels.clawdtalk]
enabled = true
telnyx_api_key = "..."
connection_id = "..."              # Telnyx SIP connection
model_provider = "voice-brain"     # references [providers.models.voice-brain]
voice = "Polly.Joanna-Neural"      # Telnyx voice ID

Full-duplex SIP voice powered by Telnyx. The agent talks over a real phone call — inbound or outbound. Supports barge-in, mid-turn tool use, and regional number provisioning.

Pair with: a telnyx provider for the brain (crates/zeroclaw-providers/src/telnyx.rs) and ensure your Telnyx account has a SIP connection with the correct webhook URL pointed at the ZeroClaw gateway.

Voice Call (Twilio / Telnyx / Plivo)

[channels.voice_call]
enabled = true
carrier = "twilio"                 # or "telnyx", "plivo"
account_sid = "..."
auth_token = "..."
from_number = "+14155550123"
stt_provider = "whisper"
tts_provider = "elevenlabs"

Traditional carrier voice — the agent picks up, transcribes with STT, replies with TTS. Higher latency than ClawdTalk but works with any regular phone number and doesn't require SIP trunk provisioning.

Voice Wake (local wake-word)

[channels.voice_wake]
enabled = true
wake_phrase = "hey claw"
engine = "porcupine"               # or "openwakeword"
model_path = "~/.zeroclaw/wake/hey-claw.ppn"
audio_device = "default"

Runs locally, listens on the mic, triggers agent interaction when it hears the wake phrase. Useful for:

• Physical voice assistants on SBCs
• Desktop "hotword → ask" workflows
• Always-listening home-automation agents

The agent doesn't send audio anywhere — wake detection is local. Only post-wake speech goes through STT and reaches the LLM.

TTS (outbound speech synthesis)

[channels.tts]
enabled = true
engine = "piper"                   # local, free
voice = "en_US-lessac-medium"
output_device = "default"

Other engines:

[channels.tts]
engine = "openai"                  # TTS-1 or TTS-1-HD
voice = "alloy"
api_key = "..."

[channels.tts]
engine = "elevenlabs"
voice_id = "21m00Tcm4TlvDq8ikWAM"
api_key = "..."

[channels.tts]
engine = "google-cloud"
voice = "en-US-Neural2-J"
credentials_json = "~/.zeroclaw/gcp.json"

[channels.tts]
engine = "edge"                    # Microsoft Edge TTS — free, online
voice = "en-US-AndrewNeural"

TTS is output-only — it's not a conversation channel; it's a speaker for other channels' replies. Pair with voice_wake for a complete local voice assistant.

Latency budget

Speech feels real-time below ~500 ms end-to-end. Practical budgets:

ClawdTalk shortcuts several of these by keeping the audio stream live; regular voice_call incurs STT + LLM + TTS sequentially.

STT options

The runtime picks STT based on stt_provider:

• whisper — local whisper.cpp; CPU-only works but GPU helps
• openai-whisper — hosted Whisper API
• deepgram — real-time streaming STT (fastest)
• google-cloud-stt
• azure-stt

Set in the voice channel's config (stt_provider = "deepgram").

Hardware notes

For always-on voice on an SBC:

• USB mic: any UAC-compliant mic works. arecord -l to verify the OS sees it
• Speaker: either USB audio out or the SBC's onboard jack; set output_device to the ALSA name
• Microphones with built-in AEC (acoustic echo cancellation) dramatically improve wake reliability when the speaker is nearby

See Hardware → Android for Android-specific audio setup.