Silence Detect

The Silence Detect node finds the silent spans in a recording or a video and returns them as a list of time ranges. Use the ranges to cut dead air from a podcast or an interview, for example with Edit Plan. The node runs locally, with no AI model and no provider key, so it works on every edition.

When to use it

• Find dead air and long pauses in a podcast or an interview before you edit it.
• Give an edit step the exact silences to remove.
• Measure how much of a recording is silence and how much is speech.

Quick start

Input and output

The output, Silences, is data, not audio. Wire it into a node that reads data, such as the Silence input of Edit Plan or an Extract Field node.

Settings

Read the result

The Silences output looks like this:

{
  "version": 1,
  "ranges": [
    { "startMs": 2120, "endMs": 3880 }
  ],
  "durationMs": 6000
}

• ranges — the silent spans, in the order they occur, in milliseconds on the clock of the original recording. endMs is exclusive and always greater than startMs.
• durationMs — the length of the whole recording, in milliseconds.

Example

A 6-second recording is silent between the 2-second and the 4-second marks. With the default settings:

1. The node finds one silent span from about 2,000 to about 4,000 milliseconds. At 2 seconds, it is longer than the 700 ms minimum.
2. The 120 ms speech padding shrinks the span at both ends. The start becomes 2,000 + 120 = 2,120 ms, and the end becomes 4,000 - 120 = 3,880 ms.

The result is the JSON above, and the run costs 11 credits.

Credits

Silence Detect costs a flat 11 credits per run. The length of the recording, the settings and the number of ranges found do not change the price.

The node analyzes a small, low-resolution copy of the sound track, so even a long recording is scanned in seconds. The first run on a recording makes that copy from the original, once. Every later run, and Transcribe on the same recording, reads the stored copy.

Tips

• Ignore breaths. Raise Minimum silence to skip short breaths and the gaps between words. Lower it to catch brief pauses.
• Protect the words. If cuts clip the start of words, raise Speech padding. If the silences feel too generous, lower it.
• Treat room tone as silence. A noisy room, for example with a fan or air conditioning, can be louder than the default threshold. Raise Silence threshold, for example from -35 to -30. Lower it, for example to -45, when quiet speech is marked as silence.
• Keep the source clock. The ranges use the original recording's timeline, so they line up with the same audio or video in a later edit.

From the API

POST /v1/silence-detect takes audioUrl and the optional thresholdDb (default -35, from -90 to 0), minSilenceMs (default 700, from 1 to 600,000) and padMs (default 120, from 0 to 60,000). The API accepts wider ranges than the settings panel. The finished run's output_data.json is the result above. The MCP tool is silence_detect. See Voice and media.