Detect Silence in Audio with Waveform Analysis: Velocaption

Detect Silence in Audio with Waveform Analysis

The silence detector scans your mic audio track to find segments below a configurable volume threshold, then highlights every silent region on the waveform so you can review and cut them.

How It Works

1. The mic track drives detection. When you open a file, the app extracts the mic audio and runs the analysis. It uses the mic track specifically because it captures speech most reliably: system audio or screen audio would trigger false positives from music, notifications, or game sounds.

2. Silent regions appear on the waveform. Every section that falls below the threshold is shaded on the waveform. Speech stays bright, silence is muted. You can see at a glance where the pauses are.

3. Adjust detection settings. Open the settings panel to fine-tune three parameters:
   • Threshold: How quiet counts as “silence.” Lower values catch only dead silence; higher values catch soft breaths and hesitations too.
   • Minimum duration: Ignore silences shorter than this. Set to 300ms to skip natural micro-pauses; set to 100ms for aggressive cutting.
   • Padding: Keep this many milliseconds of silence before and after each speech segment. Prevents words from sounding clipped at the edges.

4. Re-analyze after changes. Click the re-analyze button to re-run detection with your new settings. The waveform updates immediately: regions that were marked as silence may become speech (or vice versa) depending on your adjustments.

5. Multi-source recordings stay in sync. For bundle projects from the recorder (screen + webcam + mic), the mic track drives detection but cuts apply to all sources equally. The screen and webcam clips are trimmed at the same boundaries.

Key Details

Start conservative, then tighten. Begin with a lower threshold and longer minimum duration. Review the results, then increase the threshold or shorten the minimum duration to catch more pauses. It is easier to cut more than to restore cuts you did not intend.

Padding prevents clipped words. Without padding, cuts happen at the exact silence boundary: which can clip the leading consonant of the next word or the trailing breath of the last. 100-200ms of padding gives a natural result.

Detection is non-destructive. Changing settings and re-analyzing does not alter your source file. Every detected region is a soft marker on the timeline. You can re-run detection as many times as needed.

What’s Next

• Editing Clips on the Timeline: Manually adjust, delete, or restore detected clips
• One-Click Denoise: Strip background noise before detection for cleaner results
• Silence Remover Overview: Back to the full feature overview