Silence detection limitations
This release does not automatically trim silence. Reliable suggestions require an explicit RMS or peak threshold, a minimum duration, noise-floor guidance, and confirmation. Rather than label room tone or a quiet musical entrance as silence, the page lets you inspect the waveform, listen, and set the boundary yourself.
No middle pauses are removed automatically. Noisy recordings may never fall below a simple threshold, and compressed ambience may look continuously active. Manual boundaries remain predictable across speech, podcasts, and music. Future silence assistance would need separate tests for all-silence, short, noisy, and no-silence inputs before being exposed.
Re-encoding and export quality
MP3, AAC/M4A, AAC-ADTS, and OGG outputs are lossy encodes. Re-encoding an already lossy source can add generation loss; a higher bitrate cannot recover discarded detail. WAV writes PCM and can be much larger. FLAC preserves the newly decoded samples losslessly but does not undo loss already present in the source.
The page does not make a universal “lossless trimming” claim. Stream copy has codec-frame and container-timestamp restrictions and can move a cut to a nearby key boundary. The default re-encode makes the requested timeline predictable, applies fades when selected, and lets ffprobe verify actual duration, codec, rate, channels, and size.
Voice, podcast and music examples
For voice notes, remove setup noise at the beginning and dead air at the end, then export a mono delivery copy if spatial information is unnecessary. For a podcast excerpt, keep a precisely reviewed segment and add a short fade only when it does not cut speech. For music, preserve intentional attacks and reverb tails and keep a master before editing.
Remove-selection can delete a false start in the middle, but the joined waveform may click if the two boundaries have incompatible amplitude or phase. A small fade at the full output edges does not crossfade the internal join. For critical music editing, use a desktop editor with sample-level crossfades and keep this tool for straightforward browser cuts.
Privacy and browser performance
Inspection, waveform decoding, trimming, joining, fades, encoding, and output verification run inside the browser through the lazy-loaded FFmpeg Worker and a small peak-summary Worker. The source, filename, tags, binary content, and peaks are not uploaded or written to analytics, LocalStorage, or IndexedDB. Refreshing discards the selected file.
Large files need transient input, decoded, encoded, and output memory. The shared 500 MiB hard limit and lower mobile warning apply. Cancellation terminates the engine, tracked virtual files are deleted, object URLs are revoked on restart and page exit, and ArrayBuffer references are released. A desktop and a shorter source are safer when memory is limited.