Remove silence from audio. Keep a little breathing room.

Choose Try a synthetic pause example above. It loads an existing original four-second mono WAV and sets a −24 dBFS threshold, a 0.5-second minimum pause and 0.2 seconds kept. Then prepare the result. A quiet high tone plays continuously beneath two louder low-tone bursts, so the gaps contain sound rather than perfect digital silence.

With these example settings, quiet gaps are shortened, while the high tone remains during the louder bursts. This demonstrates the level-and-duration rule, not speech recognition, noise separation or the quality of an AI result. The usual −50 dBFS default does not find qualifying pauses in this source because the high tone keeps interrupting its quiet runs. The result you hear is generated by the same processor used for your files.

Loading an example changes the file and detection settings. It does not automatically process, play, or download audio. Keep any result you need before changing files. The source's existing generation and CC0 notes remain available; this tool does not assign a new license or claim that the older gated demonstration uses silence removal.

The detector compares the absolute decoded sample level with the threshold. A run counts as quiet only when every sample frame stays below that level for at least the minimum duration. In stereo, either channel can protect the interval from removal. The tool never averages opposing stereo signals to decide whether a frame is quiet.

Only the middle of a qualifying internal pause is removed, retaining the requested breathing room beside each wanted passage. For a long leading or trailing quiet run, the keep time stays beside the recording’s first or last above-threshold sound. Shorter pauses remain. This is a simple rule you can inspect, not a model that knows where a sentence or musical phrase should end.

A fan, music bed, or recurring click may keep the signal above the threshold and prevent cuts. Raising the threshold may find more gaps but can remove quiet speech or decays. Lower it again if wanted content disappears. If the whole recording falls below the threshold, the tool stops rather than claiming an empty file is a successful cleanup.

For desktop editing, Audacity’s Truncate Silence reference explains the same general distinction between detecting a sustained quiet passage and choosing how much time to retain. This browser tool uses the explicit linked sample-level rule above; it is not a reproduction or benchmark of Audacity’s complete algorithm.

Every removal changes the time position of everything after it. Do not replace audio under an unchanged video or beside unedited microphone tracks and expect sync to survive. Use a timeline editor to coordinate picture, audio, subtitles and other tracks, or keep the original timing.

Both channels share the same removals. Newly cut edges receive fades of up to 5 ms, shortened further for very small retained segments. The original outer edges are unchanged unless they were cut. There is no overlapping crossfade, normalization, pitch or speed change. Read the cut-boundary guide if a transition still clicks.

The browser decodes at 48 kHz. WAV writes 16-bit PCM; MP3 applies a new lossy encode at 128, 192, or 320 kbps and may add delay or padding. Values beyond integer PCM range are clipped. Original metadata and bit depth are not preserved. The duration shown for processed frames is not an independent measurement of MP3 encoding padding.

The result reports removed intervals in original decoded time, remaining duration and actual file bytes. Cutting pauses can also change integrated loudness; use the LUFS meter on the downloaded copy when that matters. Keep your original because changing the threshold cannot restore material already discarded from an exported file.