Private browser-based audio compression

Audio Trimmer Online

Trim the beginning, end, or a selected range with a responsive waveform summary, exact time inputs, keyboard-adjustable handles, optional fades, and local export. Keep the selection or correctly join both sides after removing it, then preview a newly probed result without uploading the source or waveform.

Local processing: audio, filenames, tags and waveform data are not uploaded or added to analytics.

StatusChoose audio file

Preview result

Actual output size
Actual output duration
Format
Average bitrate
Sample rate
Channels
Output timeline start
0:00
Output timeline end
Download

If this browser cannot preview the selected format, download it or choose MP3/WAV.

Trimming vs compressing

Trimming changes duration by keeping or removing a time range. File compression changes the way the remaining audio is encoded, such as bitrate, codec, sample rate, or channels. A shorter file often uses fewer bytes because it contains fewer seconds, but trimming does not automatically choose a lower bitrate or apply a dynamic-range compressor effect.

This page focuses on timeline edits and a predictable re-encoded export. The main Audio Compressor focuses on file-size settings. Keeping those roles separate makes estimates and limitations visible. The original file is never overwritten; the result is a new browser-local Blob with a sanitized download name.

How removing duration reduces size

For a roughly fixed bitrate, output size scales with remaining duration. Keeping half of a 128 kbps program produces an audio payload near half the original program time at that rate, plus container overhead. It may not be half the source bytes if the source used another bitrate, codec, channel count, VBR distribution, or large metadata.

The page estimates from selected duration and output settings, then probes actual bytes and duration after export. Remove-selection mode trims the left and right segments, resets timestamps, and concatenates them in the FFmpeg filter graph. It is offered because the underlying path performs a real join rather than merely skipping playback.

Selecting precise start and end times

The two range handles provide pointer and keyboard control, while numeric fields accept exact seconds. Start must remain before end and inside the measured duration. The selection summary reports selected and removed time. Play Selection seeks the original preview to the start and stops at the end, subject to normal browser media timing rather than sample-editor precision.

The waveform is an overview of peak buckets calculated from low-rate decoded PCM in a Worker. It does not store every original sample in Astro or React state. Canvas rendering accounts for device pixel ratio, and resizing redraws the existing peaks without decoding again. Text start, end, selection, and result values remain available without relying on the graphic.

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.

FAQ

Does trimming reduce file size?

Usually when duration is removed, but actual size also depends on output bitrate, codec, channels, metadata, and container overhead.

Can I remove a middle section?

Yes. Remove Selection trims both sides, resets timestamps, and joins them through the audio filter graph.

Is the cut lossless and sample-accurate?

No universal claim is made. The default re-encodes for predictable timing, and browser preview controls are not a sample editor.

Does the tool detect silence automatically?

No. This release keeps boundaries manual rather than applying an unexplained threshold to noise or quiet content.

Why can a lossy export sound different?

MP3, AAC, and Vorbis are newly encoded, so an already-lossy source can gain another generation of loss.

Is the waveform uploaded?

No. A bounded peak summary is calculated locally and discarded with the page task.