Wake word works once then stops hearing “hey henry” (Windows + OpenClaw)
Body:
Hi Krill team
I need help stabilizing local voice wake/command flow on Windows.
Goal
Reliable loop:
1. say “hey henry”
2. assistant wakes
3. command executes
4. can wake again reliably
Problem
• Sometimes first wake works, then “hey henry” no longer triggers
• Sometimes it hears commands incorrectly (text doesn’t match spoken words)
• Behavior is inconsistent between turns
Environment
• Windows 11 (10.0.26100)
• OpenClaw 2026.3.2
• Custom Python wake listener in workspace
• Vosk + sounddevice for wake path
• WSL2 Ubuntu available
• faster-whisper available in WSL (small.en)
• Headset mic: AW920H Chat
What we already tried
• Killed duplicate listener processes / relaunched clean
• Forced mic selection preference to WASAPI device
• Changed cooldowns and listen windows
• Reset recognizer state between turns
• Enabled conversation window after wake
• Added software input gain boost
• Switched command transcription to WSL Whisper for better accuracy
• Still seeing unreliable re-wake + intermittent missed trigger
Request
What is the recommended stable architecture for this on OpenClaw/Windows?
• Separate always-on wake process + separate command STT worker?
• Reinitialize recognizer/stream every turn?
• Known-good host API/device settings for repeat wake reliability?
• Any reference implementation pattern we should follow?
I can share the exact script if needed.
Hi Krill team
I need help stabilizing local voice wake/command flow on Windows.Goal
Reliable loop:
1. say “hey henry”
2. assistant wakes
3. command executes
4. can wake again reliably
Problem
• Sometimes first wake works, then “hey henry” no longer triggers
• Sometimes it hears commands incorrectly (text doesn’t match spoken words)
• Behavior is inconsistent between turns
Environment
• Windows 11 (10.0.26100)
• OpenClaw 2026.3.2
• Custom Python wake listener in workspace
• Vosk + sounddevice for wake path
• WSL2 Ubuntu available
• faster-whisper available in WSL (small.en)
• Headset mic: AW920H Chat
What we already tried
• Killed duplicate listener processes / relaunched clean
• Forced mic selection preference to WASAPI device
• Changed cooldowns and listen windows
• Reset recognizer state between turns
• Enabled conversation window after wake
• Added software input gain boost
• Switched command transcription to WSL Whisper for better accuracy
• Still seeing unreliable re-wake + intermittent missed trigger
Request
What is the recommended stable architecture for this on OpenClaw/Windows?
• Separate always-on wake process + separate command STT worker?
• Reinitialize recognizer/stream every turn?
• Known-good host API/device settings for repeat wake reliability?
• Any reference implementation pattern we should follow?
I can share the exact script if needed.
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
130,529Members
Resources
Similar Threads
Was this page helpful?
