About Audio Trimmer

Who we are

Audio Trimmer is a small, focused project building one thing well: a kit of honest audio tools — trimming, extracting, muting, vocal and background-music separation — that run on your own hardware instead of someone's server. The project started from a frustration every video and audio worker knows: you need one small job done (pull the sound out of a clip, cut a ringtone, kill a voice from a track), and the search results hand you a page that wants an upload, an account, a queue and maybe a credit card.

We built the opposite of that. The first screen of every Audio Trimmer page is a working tool, not a marketing video. The processing happens in your browser tab for the light jobs, and in a free, open-source Windows application on your own PC for the heavy separation jobs. Nothing you process here is uploaded, because there is no upload path in the code at all.

What we build

The product has three layers, all sharing one philosophy:

• Fourteen focused tool pages. Each targets one real search task — from trimming audio and extracting sound from video, through removing vocals from a song and making instrumentals, to removing background music from a video while keeping the speech. One job, one page, one default workflow. The pages differ where the tasks genuinely differ: the trimmer shows a real waveform, the extractors list the actual audio tracks in your container, the mute tool copies your video stream untouched, and the separation pages preview your file before any model runs.
• The browser tools. Files up to 100 MB are processed locally via WebAssembly — a build of FFmpeg compiled for the browser, loaded on demand — and the browser's own Web Audio API. Decoding, cutting, demuxing and encoding all happen on your CPU, in your tab, offline-capable after the page loads.
• The Windows separation engine. Vocal and music separation needs more compute than a browser tab can honestly deliver today. Rather than fake a progress bar, those pages hand off to a free, open-source Windows application (Ultimate Vocal Remover, MIT licence) that runs the models locally, with GPU acceleration when available and no length limit. The browser step — checking the file, previewing it, extracting the audio track — still runs right on the page.

How the pipeline works

There are two honest routes, and every page tells you which one you are on:

1. Browser-only (E0). Trim, extract, get and mute jobs run entirely in your tab. FFmpeg in WebAssembly reads the container, maps the stream you chose, copies or encodes once, and hands you the result as a normal file. The video stream is never re-rendered when it does not need to be — the iPhone mute, for example, is a stream copy, which is why the picture quality cannot change.
2. Local separation (E1/E2). Vocal removal, instrumentals, voice isolation and background-music removal need a trained model. The browser part verifies and previews your file and can extract the full audio track; the model itself runs in the recommended Windows application on your machine. Two stems come out for songs (vocals and instrumental), and a speech/music/effects split underlies the background-music workflow.

We say "speech, music and effects" carefully: those are different stems, and confusing them is how tools end up deleting dialogue while keeping the music. The homepage removes the music and keeps the people talking; the mute page removes everything; the extract pages keep everything. The words on each page mean exactly what they say.

The principles we publish

• The tool is the first screen. You came to do a job; the job is at the top of the page. The explanations, guides and FAQs live below the tool, written for people and crawlers alike, because a page that cannot explain itself cannot be trusted either.
• Honest limits over success badges. Separation leaves residue on dense mixes. Reverb ghosts. Mono recordings are harder than stereo. Browser jobs cap at 100 MB. Every page carries its limits section, because a tool that never tells you what it cannot do is marketing, not software.
• Privacy by absence. No uploads, no accounts, no analytics, no media database. The privacy policy is short because the inventory of data we could mishandle is nearly empty.
• No fake functionality. Where the browser cannot do the job, the page says so and hands you the real path. You will not find an "online AI remover" here that is secretly a queue on a server — or worse, a spinner around nothing.
• Rights stay yours to respect. Separating a vocal or muting a clip changes the audio, not the law. Every page reminds you to process what you own or are licensed to, and nothing here is a licence to infringe.

Who pays for this

The site is static and cheap to run, the browser does the computing you would otherwise rent, and the Windows engine is maintained by its open-source community. That is the whole business model: costs stay near zero, so the tools stay free. If that ever changes, the about page will be the first to say it — in the same plain words as everything else here.

Where this is going

Near-term: more of the separation moving into the browser as WebAssembly and WebGPU mature, more input formats, and deeper guides for each job. The philosophy stays: one job per page, real tools, real limits, and your files staying yours from the first click to the download.

How to judge us

The best way to evaluate a tools site is to watch what it does when nobody is processing anything. Reload a Audio Trimmer page with your network tab open: after the static assets, the only third-party request is the public CDN that serves the processing engine and the GitHub API call that resolves the installer button. No trackers, no beacons, no upload calls waiting for your file. Then drop a file into any tool and watch again: nothing leaves. We would rather be checkable than persuasive — the whole product fits in one browser tab, and one tab is easy to audit.