Live Monitoring
How BarkCard listens, classifies, and records while you're away.
How Detection Works
BarkCard captures audio from your microphone and processes it in 1-second chunks. Each chunk is fed through Google's YAMNet model — a pre-trained audio classification neural network that runs entirely on your device via TensorFlow.js.
YAMNet recognizes over 500 sound types. BarkCard focuses on 7 dog-related classes: dog (generic), bark, yip, howl, bow-wow, growl, and whimper. When any of these exceeds the sensitivity threshold, a vocalization event is logged.
Detection Sensitivity
The sensitivity slider controls how confident the AI must be before logging a detection. Lower values (more sensitive) catch quieter or ambiguous sounds but may produce false positives. Higher values (less sensitive) only log clear, confident detections.
Default: 0.30 — good balance for most environments
0.15 - 0.25 — use in quiet environments to catch soft whimpers
0.35 - 0.50 — use in noisy environments to reduce false positives
Classification Scores
Each detection stores the AI's confidence scores for all vocalization types. You can view these by clicking on any event in the report card's detection log. The scores show how confident the model was about each possible classification — for example, a sound might score 72% bark and 35% howl.
Quiet Streak
The quiet streak timer shows how long it's been since the last vocalization. This is one of the most useful metrics for tracking separation anxiety progress — longer quiet streaks mean your dog is learning to settle.
Video Recording
BarkCard continuously records video and audio from your camera throughout the session. When the session ends, it extracts short clips around each vocalization event — with padding before and after so you see the full context, not just the moment of the bark.
These clips are then stitched into a highlight reel — a single video of every vocalization moment.
Audio Level Meter
The audio level meter gives you a real-time view of the microphone input. It's useful for verifying the mic is working and seeing ambient noise levels before leaving. The meter updates continuously regardless of whether a vocalization is detected.