ROS 2 Lowwi Wakeword Detection Node
A ROS 2 node that listens to audio input and detects custom wakewords using the Lowwi library. Triggers a callback and publishes detection results when a wakeword is detected.
✅ Topics
| Topic |
Type |
Description |
/audio_stamped (default) |
audio_tools/msg/AudioDataStamped |
Raw audio input in S16LE format |
/lowwi_ww (default) |
lowwi/msg/WakeWord |
Published when a wakeword is detected |
⚙️ Parameters
| Parameter |
Type |
Description |
audio_topic |
string |
Topic to subscribe to for audio input (default: /audio_stamped) |
output_topic |
string |
Topic to publish wakeword detection messages (default: /lowwi_ww) |
wakeword.phrases |
string[] |
List of phrases to detect (e.g. ["Hey Jarvis"]) |
wakeword.models |
string[] |
Path to corresponding ONNX model files |
wakeword.min_activations |
int[] (optional) |
Minimum activations to trigger wakeword (default: 5 if omitted) |
wakeword.refractory |
int[] (optional) |
Refractory period (cooldown) per wakeword (default: 20) |
wakeword.threshold |
float[] (optional) |
Confidence threshold per wakeword (default: 0.5) |
🧩 Implementation Notes
- Written in C++ using
rclcpp.
- Subscribes to incoming audio with S16LE at 16kHz
- Uses CLFML::LOWWI to manage wakeword detection.
- Wakewords and model settings are loaded at runtime via parameters.
- Gracefully handles missing optional parameters by applying sensible defaults.
📤 WakeWord Message Structure
Published to /lowwi_ww (or a user-defined topic), of type lowwi/msg/WakeWord:
| Field |
Type |
Description |
header.stamp |
builtin_interfaces/Time |
Timestamp of the triggering audio sample |
header.frame_id |
string |
Frame from which the audio was captured |
wakeword_detected |
bool |
Always true when message is published |
wakeword_name |
string |
Name/phrase of the detected wakeword |
wakeword_confidence |
float32 |
Detection confidence (0.0 – 1.0) |
🏁 Run Example
ros2 run lowwi lowwi_node \
--ros-args \
-p wakeword.phrases:="['Hey Mycroft', 'Hey Jarvis']" \
-p wakeword.models:="['models/hey_mycroft.onnx', 'models/hey_jarvis.onnx']" \
-p wakeword.min_activations:="[2, 3]" \
-p wakeword.refractory:="[25, 30]" \
-p wakeword.threshold:="[0.6, 0.65]" \
-p audio_topic:="/my_custom_audio" \
-p output_topic:="/custom_wakeword_output"