OpenClaw Hybrid Assistant

A lightweight Linux voice channel for OpenClaw. Audio is processed locally for voice activity detection, speech recognition, and speech synthesis; requests are sent to OpenClaw over WebSocket. No local LLM is included.

Pipeline

text

microphone -> VAD -> STT -> OpenClaw WebSocket -> TTS -> speaker

The playground uses the public RunAnywhere component APIs with the Sherpa backend:

Silero VAD
Parakeet TDT-CTC 110M EN int8 by default, or Whisper Tiny EN
Piper Lessac Medium by default, or Kokoro v0.19

Listening is continuous and gated by VAD.

Wake-word activation is not supported: RunAnywhere Commons does not ship an executable wake-word backend.

Requirements

Linux with ALSA
CMake 3.16 or newer
A C++20 compiler
curl, tar, and bzip2
sox to generate the acknowledgment sound
A running OpenClaw WebSocket gateway

Build

From the repository root:

bash

bash sdk/runanywhere-commons/scripts/linux/download-sherpa-onnx.sh
bash sdk/runanywhere-commons/scripts/build-linux.sh

cd Playground/openclaw-hybrid-assistant
./scripts/download-models.sh
./build.sh

Alternative model downloads:

bash

./scripts/download-models.sh --whisper
./scripts/download-models.sh --kokoro
./scripts/download-models.sh --whisper --kokoro

Models are stored under ~/.local/share/runanywhere/Models/Sherpa/. The acknowledgment sound is stored under ~/.local/share/runanywhere/Models/ONNX/earcon/.

Run

bash

./build/openclaw-assistant
./build/openclaw-assistant --openclaw-url ws://openclaw-host:8082

Available options:

text

--list-devices
--input <device>
--output <device>
--openclaw-url <url>
--device-id <id>
--debug-vad
--debug-stt
--debug-audio
--help

Use Ctrl+C to stop the assistant.

Tests

The build produces two test executables:

bash

./build/test-components --run-all
./build/test-integration --run-all

Component test options:

text

--test-vad-stt <wav>
--test-full <wav>
--test-noise <wav>
--run-all

Integration test options:

text

--test-tts-queue
--test-chime
--test-interruption
--test-sanitization
--test-tts
--test-openclaw-flow [--delay <seconds>]
--run-all

See tests/README.md for the test matrix.

systemd

scripts/openclaw-voice.service is an example user service. Update its repository path and gateway URL before installation.

bash

mkdir -p ~/.config/systemd/user
cp scripts/openclaw-voice.service ~/.config/systemd/user/
systemctl --user daemon-reload
systemctl --user enable --now openclaw-voice.service

Troubleshooting

Missing native libraries: add sdk/runanywhere-commons/dist/linux/lib and the Sherpa library directory to LD_LIBRARY_PATH.
Missing models: rerun ./scripts/download-models.sh.
No microphone input: run ./build/openclaw-assistant --list-devices and select a device with --input.
Cannot reach OpenClaw: verify the gateway URL and that its WebSocket port is reachable.