Audio trimmer for precise local cuts

How this audio trimmer chooses the cut

A timestamp is a convenient instruction, but digital audio is a sequence of discrete samples. This audio trimmer decodes the selected file, reads its real sample rate, and maps both typed times to the nearest available sample. The start sample is included and the end sample is excluded. That half-open rule makes the output predictable: adjacent ranges do not duplicate a boundary sample, and removing a range does not leave its last sample behind.

The waveform is measured from the same decoded channels used for the edit. Each vertical mark represents the largest absolute sample in one consecutive time window across both channels. Peaks therefore stay in their real left-to-right order, including short transients that an average could hide. Use the two range controls for a quick selection, then type seconds, mm:ss, or hh:mm:ss when a precise boundary matters. The receipt prints the applied sample indexes so the rounded result is visible rather than implied.

Keep a range or remove one section

Keep selected range makes one clip from the chosen start through the chosen end. Remove selected range does the complementary job: it keeps everything before the start, keeps everything after the end, and joins those two parts in their original order. The second mode is useful for removing a false start, interruption, long pause, or private sentence without asking an automatic detector to guess what counts as unwanted material.

Keeping the complete source would only re-encode it, so that no-op is blocked until at least one edge moves. Removing the complete source is blocked because it would leave no playable audio. An empty or reversed selection is also rejected after sample rounding. Those checks happen before the worker loads the encoder. You can play the selected section from the source first, change either time, and rerun without choosing or decoding the file again.

Why the result is a new MP3

WAV and MP3 are the tested, portable inputs. M4A, AAC, Ogg, FLAC, AIFF and WebM can also work when this browser and operating system expose the matching decoder. A small signature check distinguishes common containers and stops a renamed PDF or image before Web Audio. Unknown audio is still allowed to reach the real decoder because a filename table cannot honestly predict every codec installed on a visitor’s device.

Output is deliberately one format: a new 128 kbps MP3. An arbitrary cut through compressed MP3 frames cannot preserve every source byte, so the selected PCM samples are encoded again. This is a lossy generation even when the input was already MP3. The completed bytes are opened by Web Audio once more before the result player and download appear. That final check also measures the file’s actual decoded duration and sample rate instead of copying desired values into a receipt and calling them proof.

What a cut-edge fade does

A cut can click when the waveform jumps abruptly at its new edge. Cut-edge fade is an explicit number from 0 to 5,000 milliseconds. Zero is the default and changes no sample levels before encoding. For a kept middle range, a positive value fades in at the new beginning and fades out at the new end. For a removed middle range, the retained left side fades to zero and the retained right side rises from zero at the join.

This is not a crossfade. The two retained parts never overlap, and the planned duration does not shrink. If a requested fade cannot fit beside a cut, the page asks for a shorter value rather than silently changing it. Original beginning or ending edges are not faded merely because they happen to remain in the result. The control only changes an edge exposed by the requested cut.

Large recordings use local memory

Compressed audio expands substantially when decoded to floating-point samples. The page shows the shared memory advice before decode and never rejects a file merely because of byte count. Reading and channel retention happen in bounded pieces with the browser given time between them, but the complete decoded recording and one transferred worker copy still need memory. The prepared worker keeps that copy so changing a start, end, operation, or fade does not require another file read or decode.

During export, the worker passes direct views of the retained samples to the MP3 encoder. It creates a bounded temporary chunk only where a fade must change sample levels; it does not allocate a second complete PCM track. Progress counts actual output samples encoded, not a timer. Cancel marks the current job between those chunks while leaving the prepared source available for another attempt. A memory-constrained phone can still run out of tab memory, so close other heavy tabs before starting an unusually long recording.

What stays local — and what metadata does not

The selected file, decoded samples, waveform, cut plan, temporary chunks, MP3 bytes and blob URLs stay on this device. The small LAME WebAssembly encoder is loaded from CreatorValet’s own origin only after the trim button is pressed. That request fetches the same program for every visitor; it contains no audio, filename or selected timestamps. There is no upload fallback, analytics beacon or third-party conversion service behind a codec error.

The new MP3 carries the edited primary audio only. Source tags, embedded cover art and other file metadata are not copied. Mono and stereo are supported; surround material is declined rather than downmixed without permission. For automatic pause detection, use the audio silence remover. To export a still amplitude graphic, use the audio waveform generator. A file with picture and sound belongs in the video trimmer, which keeps both tracks aligned. You can also follow the local-processing verification guide to inspect the network boundary yourself.