Remove "um", "uh", "like" and other filler words from a speech recording automatically. The Whisper engine transcribes; the Silero VAD model snaps the cuts to clean word boundaries.
Filler-Word Remover is under active development. We're building browser-based audio tools powered by FFmpeg WebAssembly. Check back soon!
Filler-Word Remover runs the audio editing and conversion job locally inside your browser. Remove "um", "uh", "like" and other filler words from a speech recording automatically. The Whisper engine transcribes; the Silero VAD model snaps the cuts to clean word boundaries. The work happens on your machine, the result is generated on your machine, and the page exposes the controls you need to drive it without burying them in menus.
Filler-Word Remover is shaped around the recurring needs of two audiences: language learners reviewing speech, who use it as a quick utility between bigger tools, and podcasters preparing episodes, who use it as their primary way of getting the job done. Both groups get the same defaults and the same speed.
Reach for Filler-Word Remover when you need a predictable result on a single file. The page works on the first visit, the controls are visible without a menu, and the output is delivered the moment the engine finishes.
The engine behind the page is whisper.cpp compiled to WebAssembly with the Whisper-tiny model. It reads your file in-memory and writes the result back into the browser. Supported inputs include MP3, WAV, OGG, FLAC, M4A, and AAC. For 200 MB and below the work usually completes in seconds; larger files mostly depend on how much spare RAM your device has.
The architecture is local-first by design. Once the page is loaded, you can disconnect from the network and the tool still completes the job. The processing stack — whisper.cpp compiled to WebAssembly with the Whisper-tiny model and the small UI shell wrapping it — ships with the page itself, so the tool keeps working in offline conditions, on a captive-portal Wi-Fi, or behind a corporate proxy that limits what the tab can reach.
As a workflow component, Filler-Word Remover is the part you reach for when a single, well-defined audio editing and conversion step needs to happen. It performs that step and returns a standard file you can carry into the next part of your pipeline.
The only practical limit is the 200 MB per-file ceiling, which keeps the tool responsive across a wide range of devices. Run the tool ten times in a row, run it ten thousand times — it behaves the same way and produces the same quality of result.
Filler-Word Remover is built around steady iteration on a small set of options rather than feature creep. Every additional setting attracts a slightly different audience, but a long settings panel makes the common case slower for everyone. The current controls reflect what users of the tool actually use.
When the job finishes, Filler-Word Remover hands you the result as `{name}-cleaned.wav`. Filenames are derived from your input where possible, so a quick batch of jobs leaves you with a tidy folder rather than a pile of generic "output (3)" files. Nothing is auto-saved on Favtoo's side because nothing was ever sent there.
From a product perspective, Filler-Word Remover is one of the simplest possible expressions of "do one thing well." The catalog contains dozens of related tools that each handle a slightly different audio editing and conversion task, and every one is a separate page rather than a tab inside a larger app. That separation keeps each tool fast to load and easy to bookmark.
Filler-Word Remover is built around the moment of need: a focused page you open when you have a specific task, complete the task, and close. The catalog contains many adjacent tools so the same model serves the surrounding parts of a typical audio editing and conversion workflow.
Useful patterns when working with Filler-Word Remover: keep the input file open in another tab so you can compare against the result; give the output file a descriptive name when saving so you can find it later (the default name is sensible but generic); and treat each run as independent — the tool has no concept of "history", which means you cannot accidentally pollute one job with leftovers from another.
If the result is not what you expected, the most common causes are easy to check. Confirm the input is under the 200 MB ceiling — files just above the cap fail silently because the engine refuses to allocate the buffer. Confirm the input is one of the supported formats. And if the page itself feels slow, try closing other heavy tabs to free up memory; the engine runs in your browser, so it competes for the same resources as everything else open.
If Filler-Word Remover solved your problem, sharing the page link with someone who has the same problem is the most useful thing you can do. The catalog grows mostly through word of mouth; visitors arriving through a recommendation tend to be the ones the tool serves best.
Default list: "um", "uh", "ah", "er", "like" (when used as filler), "you know", "I mean", "sort of", "kind of". You can edit the list before processing.
Whisper.cpp transcribes the audio with word-level timestamps. Filler matches are confirmed by the Silero VAD so we never cut into a word that happens to share characters with a filler.
Cuts snap to natural pause boundaries detected by Silero VAD, then a 25-ms crossfade smooths each splice. The result sounds like the speaker did not say the filler in the first place.
No. Whisper-tiny (~39 MB) and Silero VAD (~2 MB) both download once and cache; the entire pipeline runs locally.
Whisper handles 99 languages and the filler list defaults to English. For other languages, edit the list to match your language's common filler words ("eh" / "este" / "tipo" etc.).
Filler-Word Remover is a browser-only tool by design and does not expose a hosted API. The reason is the same as the privacy story: there is no Favtoo backend doing the work, so there is no service to call. If you need to script the same transformation, the underlying engine (whisper.cpp compiled to WebAssembly with the Whisper-tiny model) is open-source and can be used directly from your own code.
Inputs are capped at 200 MB per file, which keeps memory usage stable across phones, tablets and older laptops. You can run Filler-Word Remover as often as you need; every run produces a full-quality result.
Favtoo keeps no copy of your file because Favtoo never receives your file. Filler-Word Remover runs entirely in your browser, the input is held only in your tab's memory, and closing the tab discards it. There is no opt-in cloud history, no "recent jobs" panel synced to an account, and no server-side retention to configure — the architecture simply has nowhere for your file to be stored.
Filler-Word Remover runs in any modern mobile browser — Safari, Chrome, Firefox and the in-app browsers in most messaging apps all support the underlying APIs. Performance depends on the device: a recent phone handles typical inputs nearly as fast as a laptop, while older devices may take a few seconds longer near the 200 MB ceiling. The interface lays out cleanly on small screens, so you do not need to pinch-zoom to see the controls.
Filler-Word Remover is a static page running an open-source engine in your browser, so a typical corporate firewall does not get in the way as long as it allows JavaScript and WebAssembly to load from Favtoo. For teams that need to host it themselves on an internal network, the underlying engine (whisper.cpp compiled to WebAssembly with the Whisper-tiny model) is open-source and can be packaged into a private build with the same behaviour. Reach out via the Contact page if that is something you are exploring.
Filler-Word Remover is free to use. The processing runs in your browser, which keeps the per-user cost low enough that the tool can be offered openly. The download is the same file the engine produced — you can use it for as many runs as you need.
Filler-Word Remover only needs the standard web platform — file picker access for the inputs you choose to load, and optionally clipboard access if you copy the result rather than downloading it. There is no microphone, camera, geolocation or background-permission request, because none of those are needed for the work the tool does.
Compress Audio
Shrink any audio file to a smaller size by lowering the bitrate. Pick a target quality (96, 128, 192, 256, or 320 Kbps) or output format (MP3, OGG, M4A) and the file is re-encoded right inside your browser using FFmpeg WebAssembly. Nothing is uploaded — your audio never leaves your device.
Convert Audio
Convert any audio file between MP3, WAV, OGG, FLAC, M4A, AAC, and Opus right in your browser. Pick the output format and (for lossy formats) the target bitrate. Everything runs locally with FFmpeg WebAssembly — your file is never uploaded and no account is required.
Audio Recorder
Record from your microphone directly in the browser. Pick quality (high, medium, low), toggle echo cancellation, noise suppression and auto-gain, then save to WebM/Opus or M4A/AAC. Audio is captured locally — nothing is uploaded.
Text to Speech
Type or paste text, pick a system voice, and listen instantly. Adjust speaking rate (0.5×–2×), pitch, and volume in real time. Uses your browser's built-in Web Speech API — no cloud TTS, no API keys, no costs.
Tone Generator
Generate a pure tone at any frequency from 20 Hz to 20 kHz. Pick a sine, square, triangle, or sawtooth waveform, choose duration, amplitude, and mono/stereo. Exports a 16-bit PCM WAV file at 44.1 kHz with built-in click-preventing fades.
Silence Generator
Generate a perfectly silent WAV file of any length from 1 second up to 1 hour. Pick mono or stereo, get a 16-bit PCM WAV at 44.1 kHz. Useful as padding between clips, intro silence, leader audio for video timing, or test material.
White Noise Generator
Generate white, pink, or brown noise as a 16-bit PCM WAV file. Pick noise type, duration up to 1 hour, amplitude, and mono/stereo. Useful for sleep, focus, masking distractions, audio testing, and as a backing layer for ambient music.
Metronome
A precise browser-based metronome powered by the Web Audio API. Set BPM from 30 to 300, choose a time signature, accent the first beat, and use tap-tempo to sync. Click timing is sample-accurate using lookahead scheduling — much steadier than typical JavaScript setInterval beats.
