DaVinci Resolve Track Silence Cutter

CLI tool for cutting silence on a single selected audio track while preserving absolute timeline timing.

Workflow

1. In DaVinci Resolve, export the timeline as Final Cut Pro 7 XML (xmeml) or FCPXML (fcpxml).
2. Run this CLI on the exported file (see below).
3. Import the written XML back into Resolve.

What The Tool Does

• Lets you choose a target audio track (A1/A2/...).
• Detects silence per clip on that track using:
  • Silence threshold in dB
  • Minimum silence duration
  • Pre-attack / post-release context in milliseconds
• Removes silent regions inside each clip.
• Keeps global timeline spacing (no ripple shift).
• Optional: splits a chosen video track at the same cut boundaries while keeping all video segments.
• Supports dry-run preview.

Project Layout

• src/cli/runXmlSilenceCut.ts — CLI entry (XML mode).
• src/core/ — silence detection, FFmpeg extract, xmeml/fcpxml cutters.
• tests/*.test.ts — Vitest tests.

Requirements

• Node.js 20+ and pnpm (see packageManager in package.json for the pinned version).
• ffmpeg on PATH (used to decode source clip audio).

Install dependencies:

pnpm install

CLI (XML mode)

The CLI runs tsx on the TypeScript source (no tsc build required for normal use).

Single track (recommended — pnpm start runs tsx on src/cli/runXmlSilenceCut.ts). Pass flags directly after pnpm start (do not insert an extra --; pnpm can forward a literal -- and confuse the parser):

pnpm start --input "/path/in.xml" --output "/path/out.xml" --track 1 --threshold-db -40 --min-silence-ms 250 --pre-attack-ms 80 --post-release-ms 120 --min-keep-ms 120 --zero-crossing-ms 12

Or run tsx on the entry file directly (same behavior):

pnpm exec tsx src/cli/runXmlSilenceCut.ts --input "/path/in.xml" --output "/path/out.xml" --track 1 --threshold-db -40 --min-silence-ms 250 --pre-attack-ms 80 --post-release-ms 120 --min-keep-ms 120 --zero-crossing-ms 12

CLI help:

pnpm run start --help

Dry-run example:

pnpm start --input "/path/in.xml" --output "/path/out.xml" --track 1 --dry-run

Multi-track with different settings in one run:

pnpm exec tsx src/cli/runXmlSilenceCut.ts --input "/path/in.xml" --output "/path/out.xml" --track-config "1:-40:250:80:120" --track-config "2:-35:300:60:90"

--track-config format is track:thresholdDb:minSilenceMs:preAttackMs:postReleaseMs:minKeepMs. Omitted values fall back to defaults (-28:500:60:60:120 unless you change the DEFAULT_* constants in src/cli/runXmlSilenceCut.ts):

pnpm exec tsx src/cli/runXmlSilenceCut.ts --input "/path/in.xml" --output "/path/out.xml" --track-config "1:-34" --track-config "2:::6"

Optional global tuning (applies to every track run in the same command):

• --analysis-frame-ms — RMS loudness window in ms (default 10). Try 18–25 for speech so short dips at the ends of words are less likely to be called “silence”.
• --pre-attack-ms — keep extra context right before speech starts (default 60, legacy alias: --pad-before-ms).
• --post-release-ms — keep extra context right after speech ends (default 60, legacy alias: --pad-after-ms).
• --min-keep-ms — minimum non-silent island to keep (default 120). Increase to 150–250 when you see many tiny chopped blocks/noise islands.
• --zero-crossing-ms — snap each keep-segment boundary to the nearest waveform zero-crossing within this radius (default 12, set 0 to disable). Helps reduce hard boundary clicks.
• --split-video-track — optional 1-based video track index (V1=1) to split at the same positions as audio cuts. Video clips are not deleted; they are only segmented.

Example with matching video cuts:

pnpm start --input "ep10.xml" --output "ep10-cut.xml" --track 2 --split-video-track 1 --threshold-db -48 --min-silence-ms 900 --pre-attack-ms 80 --post-release-ms 200

Optional: emit JavaScript with pnpm run build if you want a compiled dist/ tree (not required for normal use).

Speech / podcast (avoid cutting tails early)

If consonants, breath, or room decay sound clipped: use a more negative --threshold-db (e.g. -48 to -55), longer --min-silence-ms (e.g. 700–1200 so only real pauses are removed), and tune --post-release-ms (e.g. 120–250) and --pre-attack-ms (e.g. 60–140) to preserve natural word boundaries. If you still see lots of tiny cut islands, raise --min-keep-ms to 180–300. Keep --zero-crossing-ms around 8–20 for cleaner boundaries. Always --dry-run first and check the summary; hundreds of tiny segments per clip usually means the detector is still too aggressive.

Example:

pnpm start --input "ep10.xml" --output "ep10-cut.xml" --analysis-frame-ms 20 --min-keep-ms 180 --zero-crossing-ms 12 --track-config "2:-50:900:80:200:180" --track-config "3:-50:900:80:200:180"

Multi-track with shorter per-track config (track:thresholdDb:minSilenceMs; other values use defaults), analysis tuning, and splitting V1 at audio cut boundaries:

pnpm start \
  --input "$HOME/Desktop/ep10.xml" \
  --output "$HOME/Desktop/ep10_cut.xml" \
  --analysis-frame-ms 20 \
  --track-config "2:-32:600" \
  --track-config "3:-36:600" \
  --zero-crossing-ms 12 \
  --split-video-track 1

Run tests:

pnpm test

Notes

• Supports FCP7 XML (xmeml) and FCPXML (fcpxml) timelines.
• If a clip has no readable source file path in XML, it is skipped with a warning.
• The CLI only modifies the selected audio track.