webcam-capture

Capture images, audio, and video from a webcam via Windows native APIs. Includes wake word listener for voice-activated photo capture. Designed for WSL2 environments where USBIP isochronous transfers fail.

Features

Mode	Script	Output	Format
Image	wincam.py	Snapshot	JPEG 640x480
Audio	win_audio.py	Recording	WAV 44.1kHz PCM mono
Video	win_video.py	Recording	MP4 H.264 + AAC
Wake Word	win_wake.py	Voice control	Listens for "DOSBot"

Wake Word Listener (Voice Control)

IMPORTANT: This script must be running for DOSBot to listen for the wake word. It continuously streams audio from the mic. When you say "DOSBot", it takes a photo automatically.

Manual Start

& 'C:\Users\doschott\AppData\Local\Programs\Python\Python312\python.exe' 'C:\Users\Public\claw-webcam-capture\listener-jarvis\win_wake.py'

• Say "DOSBot" → takes a photo to claw-webcam-capture\image\
• Uses Google Speech Recognition (requires internet)
• Press Ctrl+C to stop listening
• Action log at C:\Users\Public\claw-webcam-capture\listener-jarvis\wake_log.txt

Auto-Start with Windows Task Scheduler

To have DOSBot listening automatically on startup:

# Create scheduled task (runs at logon)
$action = New-ScheduledTaskAction -Execute "C:\Users\doschott\AppData\Local\Programs\Python\Python312\python.exe" -Argument "C:\Users\Public\claw-webcam-capture\listener-jarvis\win_wake.py" -WorkingDirectory "C:\Users\Public\claw-webcam-capture\listener-jarvis"
$trigger = New-ScheduledTaskTrigger -AtLogOn
$settings = New-ScheduledTaskSettingsSet -AllowStartIfOnBatteries -DontStopIfGoingOnBatteries
Register-ScheduledTask -TaskName "DOSBot_WakeWord" -Action $action -Trigger $trigger -Settings $settings -Description "DOSBot wake word listener"

# Start manually:
Start-ScheduledTask -TaskName "DOSBot_WakeWord"

# Stop:
Stop-ScheduledTask -TaskName "DOSBot_WakeWord" -ErrorAction SilentlyContinue

# Remove:
Unregister-ScheduledTask -TaskName "DOSBot_WakeWord" -Confirm:$false

How It Works

1. Script streams audio from Brio 100 mic continuously
2. Energy threshold detects when you're speaking
3. After silence, audio is sent to Google Speech Recognition
4. If "image" is found in the text → triggers photo capture
5. Photo saves to claw-webcam-capture\image\webcam_capture.jpg

Why This Approach

WSL2's USBIP driver (vhci_hcd) cannot handle isochronous USB transfers — the high-bandwidth, low-latency transfers webcams use for video and audio. USBIP produces placeholder images (~13KB with EXIF metadata, no real pixels). This skill uses Windows native APIs instead:

• Video/Audio: DirectShow via ffmpeg
• Image: DirectShow via OpenCV
• Audio only: WASAPI via PyAudio

Setup

Windows

winget install Python.Python.3.12
python -m pip install opencv-python-headless pyaudio speech_recognition
winget install Gyan.FFmpeg

# Create folder structure:
New-Item -ItemType Directory -Path 'C:\Users\Public\claw-webcam-capture\image' -Force
New-Item -ItemType Directory -Path 'C:\Users\Public\claw-webcam-capture\audio' -Force
New-Item -ItemType Directory -Path 'C:\Users\Public\claw-webcam-capture\video' -Force
New-Item -ItemType Directory -Path 'C:\Users\Public\claw-webcam-capture\listener-jarvis' -Force

Copy the scripts

cp webcam-capture/wincam.py /mnt/c/Users/Public/claw-webcam-capture/image/wincam.py
cp webcam-capture/win_audio.py /mnt/c/Users/Public/claw-webcam-capture/audio/win_audio.py
cp webcam-capture/win_video.py /mnt/c/Users/Public/claw-webcam-capture/video/win_video.py
cp webcam-capture/win_wake.py /mnt/c/Users/Public/claw-webcam-capture/listener-jarvis/win_wake.py

Usage

# Image capture
& 'C:\Users\doschott\AppData\Local\Programs\Python\Python312\python.exe' 'C:\Users\Public\claw-webcam-capture\image\wincam.py'

# Audio recording (5 seconds)
& 'C:\Users\doschott\AppData\Local\Programs\Python\Python312\python.exe' 'C:\Users\Public\claw-webcam-capture\audio\win_audio.py'

# Video recording with audio (5 seconds, or pass duration as arg)
& 'C:\Users\doschott\AppData\Local\Programs\Python\Python312\python.exe' 'C:\Users\Public\claw-webcam-capture\video\win_video.py' 10

# Wake word listener
& 'C:\Users\doschott\AppData\Local\Programs\Python\Python312\python.exe' 'C:\Users\Public\claw-webcam-capture\listener-jarvis\win_wake.py'

# From WSL, copy results:
cp /mnt/c/Users/Public/claw-webcam-capture/image/webcam_capture.jpg output.jpg
cp /mnt/c/Users/Public/claw-webcam-capture/audio/audio_capture.wav output.wav
cp /mnt/c/Users/Public/claw-webcam-capture/video/video.mp4 output.mp4

Technical Details

• Image: 640x480 JPEG, native YUYV mode (not forced MJPG), 5-frame warmup for auto-exposure
• Audio: Brio 100 mic as "Microphone (Brio 100)", 44.1kHz mono 16-bit PCM, ~88KB/s
• Video: ffmpeg DirectShow, H.264 video + AAC audio, ~370KB/s
• Devices: Uses Brio 100 webcam + built-in mic via DirectShow/WASAPI

Requirements

• Windows 10/11
• Python 3.12 with OpenCV, PyAudio, and SpeechRecognition
• ffmpeg (Gyan.FFmpeg) for video recording
• Logitech Brio 100 (or any DirectShow-compatible webcam/mic)