Audio Frequency Eraser
Select, reduce and restore sound on a spectrogram.
Private by design
Your inputs stay on your device and are not sent to our servers. Optional usage analytics does not receive your files or tool inputs.
How to use Audio Frequency Eraser
- Open a browser-supported audio file up to 100 MB and 10 minutes, or try the generated sound sample.
- Use Select or Brush on the frequency threads or heatmap. Switch visualizations in View. On touch devices, Expand to draw; exact time and frequency fields also work inline. Choose a channel, reduction and soft edges. Restore returns original sound in painted areas.
- Compare Original, Edited and Removed signals. Zoom, loop a selected time range, undo or redo, and use Keep selection to isolate frequency regions. Wait for the measured preview to finish before auditioning edited audio.
- Export the edited or removed signal as the preview's 44.1 kHz 16-bit WAV, or prepare MP3 at 128–320 kbps. Download before leaving the page.
What are the limits of Audio Frequency Eraser?
Local spectral editing for mono/stereo recordings up to 100 MB and 10 minutes. Browser codec support varies. Processing runs at 44.1 kHz in bounded chunks with overlapping context and a track-wide peak measurement pass. One cancellable worker is reused; a 4096-sample FFT and overlap-add reconstruction preserve chunk boundaries. Browser decoding still needs memory for the source, so longer tracks may exceed older devices' available memory. Frequency resolution is about 10.8 Hz and each analysis window spans about 93 ms; edits are not sample-precise, and sharp masks can ring or smear transients. At most 40 edits and 512 region marks; long brush strokes stop at 64 marks. Soft edges smooth transitions; time feathering is capped at 25 ms. Frequency editing affects overlapping instruments and cannot reliably isolate vocals, guitar, piano or drums. Hum/hiss/rumble presets select frequency bands, not recognized sounds. Source peak protection may reduce the decoded recording before editing, with a visible notice; output peak protection may further reduce each signal. WAV stores processed 16-bit PCM; MP3 adds lossy encoding and possible frame padding. Source metadata is not preserved. Edits are temporary; nothing is uploaded or saved to an account. The included sample is a generated test signal, not an instrument-separation example.