Free tools for your next audio edit
Reduce a known steady hum with narrow frequency cuts. Choose 50 Hz, 60 Hz or another whole-number frequency, add only the harmonics you need, and compare a real WAV or MP3 export. Free local processing, without an account or audio upload.
Targets 60 Hz. Keep the original for comparison.
No file selected. Audio stays in this browser.
Applies one to six narrow notch filters independently to each channel. No automatic noise detection, normalization, trimming, source separation or reconstruction of damaged words. Filter phase and startup transients change the waveform; peaks can rise. Download before leaving and keep your original.
A focused test for steady tonal noise
Keep the original and compare a complete phrase. Less hum is useful only if the voice or music still serves your edit.
Open mono or stereo MP3/WAV up to 50 MiB and five minutes. Include a pause, a quiet wanted phrase and an important low note. The browser decodes locally at 48 kHz; no audio is uploaded.
Enter the hum frequency—often 50 or 60 Hz, but identify your actual recording. Begin with one notch at the fundamental. Add the next multiples only if they address a remaining buzz. Higher Q makes narrower cuts that need a more accurate center.
Choose WAV or MP3 and prepare it. Compare the original and filtered players at a similar listening level. Download the same file used by the result player. Changing a setting clears that result; every new attempt starts from the selected original.
Narrow frequency cuts
This tool cascades one to six second-order notch filters. The center frequencies are the selected fundamental and successive whole-number multiples: 60 Hz with three notches means 60, 120 and 180 Hz. The allowed fundamental is 20–1,000 Hz and Q is 5–100, in whole numbers. The targeted frequencies are displayed before you process.
The same coefficient sets are used on both stereo channels, with independent double-precision state for each channel and each notch. Audio is not mixed from one side to the other. The notch equations follow the W3C Audio EQ Cookbook. Filtering is causal, starts from zero state and changes phase; it is not a zero-phase process.
A notch cannot recognize whether a note is wanted. A low singing voice, bass instrument or room sound in the same bands can change along with the hum. Do not select six notches merely because six are available. Start with the smallest useful change and compare the difficult phrase.
Match the method to the sound
Test a notch at the identified frequency, then inspect any remaining harmonics. A moving or non-integer frequency may not line up with these fixed whole-number centers. The hum-removal guide explains a spectrum-based inspection in a desktop editor.
Compare the local high-pass filter when a wider low-frequency cut fits the problem. That filter also affects wanted bass and voice body. Raising a low-cut cutoff is a different operation from targeting separate harmonics.
The local noise gate lowers quiet passages by level. A gate can leave hum beneath a louder open phrase. Use the cleanup-method guide for hiss, overlapping sources and damaged recordings.
An original repeatable comparison
The original listening examples combine known 60 Hz and 120 Hz tones with a wanted 440 Hz tone. Download the original synthetic WAV, choose it above, and set 60 Hz, two notches, Q 20 and WAV. Filtering runs only when you choose it; there is no automatic playback.
Compare your exported WAV with the guide's independently generated FFmpeg reference. The files can differ at the quantization boundary; native sample checks allow two PCM16 steps against that reference. This demonstrates known filter behavior, not speech recovery or a real microphone's noise-removal quality.
The three original guide WAVs are dedicated under CC0 1.0. Read their generation and verification notes and component measurements. Do not read the tiny remaining synthetic tone amplitudes as a 100 dB overall quality gain.
Know what you keep
WAV exports 48 kHz, 16-bit PCM with the decoded mono/stereo layout and source frame count. MP3 encodes once at 128, 192 or 320 kbps and may add codec delay or padding. Each actual output must fit 50 MiB. Tags, artwork, compressed bytes and the original bit depth are not copied.
No normalization, limiter, trimming, edge fade or appended filter tail is applied. The changed phase and transient response can raise peaks. Any samples clipped in final PCM preparation are counted. If the count is nonzero, lower the original source level in an editor and retry; lowering the already clipped download cannot recover its peaks. The count is not a true-peak or decoded MP3 measurement.
Keep the best original and download before leaving. If you are replacing a video's soundtrack, retain its source position and use the audio replacement guide to check the final synchronization. For an actual speaking voice alone, AI voice isolation is a separate source-estimation service with account and download-credit requirements.
MP3 encoder credits
The optional MP3 download uses lamejs 1.2.1 by Alex Zhukov, a JavaScript port of the LAME MP3 encoder. Its unmodified script is served separately. See the attribution and source notes, upstream license notice, LGPL-3.0 and GPL-3.0 texts, and complete upstream source package.
A few useful answers
Choose a mono or stereo MP3/WAV up to 50 MiB and five minutes, enter the actual hum frequency, and start with one notch. Choose WAV or MP3, filter, and compare the real export against the original. Add harmonics only when the remaining buzz needs them. Processing and downloads are free and local.
No. You choose a whole-number fundamental from 20 to 1,000 Hz. The tool does not analyze a noise profile or identify which sound is unwanted. A narrow notch at the wrong frequency may leave the hum, and a tone that drifts can escape a fixed cut. Use the guide's spectrum-inspection workflow when you need to identify it.
One targets the fundamental alone. Each extra notch adds the next whole-number multiple, up to six total. At 60 Hz with three notches, the cuts are at 60, 120 and 180 Hz. These cuts affect wanted sound in those bands too; more filters do not automatically mean a better repair.
Q controls narrowness, from 5 to 100 here. A higher value narrows each cut and requires a more accurate frequency. Start by comparing the displayed Q 20 and a single fundamental; change one setting at a time while checking wanted low notes and speech. No setting is guaranteed to suit every recording.
Not generally. It applies fixed narrow frequency cuts, not a broad noise profile, wind recognition, a gate or AI separation. It can change wanted notes and cannot reconstruct damaged speech. The separate high-pass filter handles a wider low-frequency cut; voice isolation is a different account-and-credit service.
WAV is 48 kHz, 16-bit PCM with the decoded source's channel and frame counts. MP3 uses one new lossy encode at 128, 192 or 320 kbps and can add delay or padding. Either actual file must fit 50 MiB. Original compressed bytes, tags, artwork and bit depth are not preserved. Preview and download use the same file.
No. There is no normalization, limiter, edge fade or appended filter tail. The filters change phase and transients, and peaks can rise. Out-of-range samples clipped during PCM preparation are counted; lower the original level and retry if needed. The count does not measure true peak or the decoded MP3 peak.
After preparing your audio