Quick answer: to remove hiss from audio without turning the wanted sound dull or metallic, separate the noise into two jobs. First, measure and reduce any stable electrical hum with narrow EQ bands or a dedicated de-hum plug-in. Then treat the remaining broadband hiss with Effects → DeHisser in Sound Forge. Set the threshold while listening to Inverse, lower the final Reduction until speech and transients return, and compare at matched loudness. A little natural noise is usually better than a silent file with watery artifacts.
If the hum exists in the recording chain, fix that before recording another take. Software can attenuate a contaminated file; it cannot turn a noisy preamp, ground loop, unbalanced cable beside a power supply, or badly aligned tape transfer into a genuinely clean source. The workflow below covers both sides: stopping the noise where possible and restoring the file when the take cannot be replaced.
This is a Windows audio-restoration guide built around the current Sound Forge tools, not a recipe of magic frequencies. It covers hiss, 50/60 Hz hum, harmonic buzz, low rumble and changing fan or room noise separately. For a broader steady-noise workflow, use the background-noise guide; for clicks and crackle, use the separate click-and-pop repair workflow.
Identify the Noise Before Removing It
Listen to a pause, then inspect a short representative selection. The waveform tells you when the noise happens. A spectrum tells you where its energy sits. Those two views prevent the most expensive restoration mistake: using a broad processor on a narrow problem.
| What you hear | What to look for | First tool | Main risk |
|---|---|---|---|
| Hiss | Broad, steady energy across many frequencies, often obvious in the upper mids and highs | DeHisser or noise-profile reduction | Dull consonants, cymbals and tape ambience |
| Hum | One narrow low-frequency line with evenly spaced harmonics | Narrow EQ cuts or dedicated De-hum | Removing bass fundamentals and body |
| Buzz | A stronger harmonic ladder that reaches higher than a soft hum | Several measured notches or De-hum | A hollow, comb-filtered result |
| Rumble | Broad low-frequency energy from handling, traffic, turntable or HVAC | High-pass filter, only as high as needed | Thinning a voice, kick, bass or piano |
| Variable fan or room noise | The spectrum and level change over time | Adaptive/spectral denoiser, automation, or re-recording | Underwater speech and moving artifacts |
| Clicks and crackle | Short impulses rather than a continuous floor | Click/pop repair | Smearing attacks if treated as hiss |
Don’t diagnose by name alone. “Static” may mean hiss, clipping, electrical buzz or intermittent digital crackle. “Hum” may have a weak 50/60 Hz fundamental but strong harmonics, and rectified power-supply noise may be more obvious near 100/120 Hz. The analyzer decides where to cut.
Fix Hiss and Hum at the Source First
Before processing a new recording, mute or disconnect one part of the chain at a time. Start at the microphone or instrument and work toward the computer. If the noise remains after an input is disconnected, the fault is later in the chain. If it disappears, reconnect the previous stage and inspect its cable, power supply and gain.
- Move audio cables away from AC adapters, dimmers and power cables. Crossing at right angles is better than running them side by side.
- Use balanced connections where the equipment supports them. Shure’s audio troubleshooting guide identifies unbalanced runs near power and lighting as a common hum pickup path.
- Test laptop power. Record briefly on battery with nonessential USB devices disconnected. If the noise changes, investigate the power or USB path rather than filtering every future take.
- Set gain at the right stage. Recording too quietly and adding large digital gain later raises the preamp and room noise with the voice. Recording too hot creates clipping, which needs the separate clipped-audio repair workflow.
- For tape or vinyl, service the transfer chain. Clean the medium and playback path, verify grounding, and check tape-head alignment before capture. Post-processing should not compensate for a preventable transfer fault.
The practical test is simple: record ten seconds of the intended setup, ten seconds with the source disconnected, and ten seconds with suspect power or USB devices removed. Compare those files before touching an effect. Shure’s noise-identification guide makes the same source-first distinction between hum, buzz, interference and other faults.
Prepare a Safe Restoration Session
- Keep the original capture. Save a working copy in WAV at the original sample rate and bit depth. Never denoise an MP3 and overwrite the only source.
- Mark three test regions. Use a quiet passage, a dense or loud passage, and a fade or word ending. One setting must survive all three.
- Protect true silence and wanted room tone. A noise-only passage is useful for analysis, but deleting every pause can make edits pump or expose hard cuts.
- Make one change at a time. Hum notch, hiss reduction, gate and compression should be auditioned separately before they become a chain.
- Level-match bypass. The quieter version often seems cleaner. Match output level closely enough that the decision is about damage, not loudness.
For a repeatable non-destructive chain, the current Sound Forge Plug-In Chainer guide explains how to order, bypass and save processors.
Measure the Noise with Spectrum Analysis
- Select a short section where the noise is exposed but representative.
- Open View → Spectrum Analysis.
- Look first below roughly 500 Hz for narrow stable lines, then scan the full range for a broad floor.
- Move the cursor to the lowest stable peak and note its actual frequency.
- Check whether additional peaks repeat at integer multiples.
- Analyze a second section. If the peaks move, a static notch may not track the problem.
The current Sound Forge Spectrum Analysis help describes the view as amplitude by frequency rather than amplitude by time. Boris FX also recommends a manageable selection in the Spectrum Graph workflow; a very long selection takes longer and hides time variation. The full Spectrum Analysis guide covers FFT size, windowing and cursor readouts when the peak is hard to isolate.
Use the spectrum to locate, then your ears to decide. A visible line is not automatically objectionable. The same frequency can belong to a bass note, vocal fundamental, transformer hum or room mode. A/B the proposed cut on the most exposed and most musically important sections.
Remove 50/60 Hz Hum and Harmonic Buzz
To remove hum from audio safely, begin with the measured fundamental. In many systems, mains-related noise starts near 50 Hz or 60 Hz and repeats at integer multiples. That is a hypothesis, not a preset. Tape-speed drift, equipment design and power-supply rectification can shift the strongest line. Measure the actual frequency before entering a number.
Method 1: narrow cuts with Modern Equalizer
- Open FX Plug-Ins → Apply Plug-In Chain and add Modern Equalizer.
- Add a peaking band at the measured fundamental.
- Use a narrow Q and lower Gain only until the tone stops dominating.
- Add a second band only when the next harmonic is clearly visible and audible.
- Bypass the EQ. If the source loses body or a musical note collapses, reduce the cut or narrow the band.
- Check the entire file. A fixed notch is appropriate only while the interference stays put.
The current Boris FX Modern Equalizer reference documents peaking filters, Frequency/Gain/Q control, up to eight bands and A/B bypass. Eight available bands do not mean eight cuts are required. Every notch removes wanted audio at the same frequency.
Method 2: a dedicated De-hum plug-in
A dedicated processor is faster when a stable fundamental creates a long harmonic ladder. Current iZotope De-hum documentation supports common 50/60 Hz bases, a free-frequency mode for shifted hum, manual or adaptive behavior and harmonic filtering. It is a separate product; availability depends on the installed and licensed VST package.
Use manual mode for a stationary tone. Use adaptive behavior only when the frequency really drifts, because a moving model has more opportunity to react to wanted bass or sustained notes. If the plug-in offers Hum Only or output-noise monitoring, listen to it. Recognizable voice, bass or music in the removed signal means the setting is too broad or too deep.
Remove Broadband Hiss with Sound Forge DeHisser
DeHisser is for steady low-level broadband noise such as tape, microphone, preamp or converter hiss. It does not require a captured noise sample. In the current Sound Forge help, open it at Effects → DeHisser.
- Select the hardest passage. Start with a quiet phrase, exposed sustain or fade where hiss competes with wanted detail.
- Open Effects → DeHisser. The current Sound Forge DeHisser page confirms the effect and path.
- Raise Reduction temporarily. This makes the Absorption boundary easier to hear; it is not the final setting.
- Increase Absorption slowly. Stop when the hiss is controlled but consonants, cymbals, pick attack and room tone remain intact.
- Enable Inverse. You should hear mostly noise. If you can identify words, melody, transients or reverb, back Absorption down.
- Lower Reduction. Restore enough residual noise that the wanted signal sounds natural in context.
- Check the other marked regions. A setting that works in a pause may chew into a loud, bright section or make a fade chirp.
- Compare bypass at matched level. Apply only after the improvement survives normal listening volume and headphones.
The detailed Boris-hosted DeHisser control reference lists Reduction from 0 to 30 dB and warns that excessive Absorption can discolor the signal or create phase-like and chirping artifacts. It specifically recommends lower reduction on very quiet material. Treat 30 dB as a control limit, not a target.
Overlap and Resolution can change the artifact/noise tradeoff and CPU use, but neither has a universally “best” maximum. Find the Absorption boundary and final Reduction first. Change advanced controls only when a repeatable problem remains.
When to Use a Noise Profile or Adaptive Denoiser
DeHisser fits uniform hiss. A noise-profile processor is more appropriate when you have a clean sample of a specific fan, air conditioner, camera whine or room bed. An adaptive processor is the fallback when the noise changes and there is no stable noise-only passage.
- Find a noise-only region from the same recording setup.
- Capture the profile from a representative section; do not duplicate one tiny block to make it “longer.”
- Apply a conservative pass to a copy or short test selection.
- Listen to the residue or removed signal.
- Lower sensitivity or reduction when wanted speech or music appears in the residue.
- Use two light passes only if each pass is independently cleaner than one aggressive pass.
The current Audacity noise-reduction manual is not a Sound Forge menu guide, but its restoration warnings are universal: constant noise is a better candidate than clicks or irregular traffic, a notch should precede broadband reduction for hum, and aggressive settings can create “musical noise” or tinkling artifacts. It also states the hard limit plainly: when noise is loud, variable or close to the wanted signal, satisfactory removal may be impossible.
Use a Different Workflow for Voice, Tape, Vinyl, and Music
Voice, podcast, and interview
Protect consonants and word endings. Reduce stable hum first, then apply light hiss reduction while checking S, F and SH sounds. A gate can lower noise between phrases, but it cannot remove hiss under speech. If the room noise rises and falls with every word, ease the denoiser and use clip gain or automation on pauses instead of forcing the gate.
Cassette and reel-to-reel transfer
Capture at a healthy level before restoration and keep the untouched transfer. Check azimuth, head cleanliness, tape speed and any original Dolby setting before assuming the high-frequency balance is “hiss.” A tape recorded with Dolby noise reduction but played without the matching decode can sound unnaturally bright; a dehisser cannot reconstruct the intended encode/decode relationship.
Sample the noise near the material you are repairing because the tape noise can change across a reel or cassette. Preserve a little stable tape bed when removing it completely would erase ambience or cymbal tails.
Vinyl transfer
Fix grounding hum before capture. Then remove isolated clicks and large pops before broadband reduction so they do not distort the noise estimate. Follow the complete Sound Forge vinyl-restoration workflow for cleaning, capture, click repair, hum, rumble and final sides.
Music or stereo master
Be cautious with narrow notches on bass notes and with hiss reduction on cymbals, strings and reverb. Process mid and side channels separately only when the evidence supports it; otherwise channel-selective work can shift the stereo image. If the noise is part of an already limited master, a small residual may be less audible than the damage caused by another aggressive process.
A Safe Processing Order
| Order | Problem | Why it comes here | Verification |
|---|---|---|---|
| 1 | Source and transfer faults | Prevents more contaminated audio and may eliminate the need for processing | Short comparison recordings |
| 2 | Severe clicks or isolated events | Prevents impulses from confusing broadband analysis | Attack and transient check |
| 3 | Stable hum or whistle | Narrow correction removes specific lines before broadband modeling | Analyzer plus bypass |
| 4 | Broadband hiss or room bed | Works on the remaining stationary floor | Inverse/residue and fades |
| 5 | Gate or pause automation | Cleans exposed gaps after the continuous noise is controlled | Phrase starts and endings |
| 6 | EQ, compression, mastering | These can exaggerate residual noise and should follow repair | Full-file A/B at matched level |
This order is a strong default, not a law. A high-pass filter may be needed before a processor that reacts badly to subsonic rumble. A click buried under heavy noise may be easier to expose after a light denoise pass. Make the exception on a copy and keep it only when the full-file A/B proves it.
Fix Metallic, Watery, or Hollow Artifacts
| Artifact | Likely cause | First correction |
|---|---|---|
| Metallic or “tinkly” noise | Noise model is wrong or reduction/sensitivity is too aggressive | Choose a better sample; lower reduction and sensitivity |
| Underwater speech | Changing noise overlaps the voice and the model is tracking both | Use less adaptive reduction; automate pauses; consider re-recording |
| Dull S sounds and cymbals | DeHisser Absorption is crossing into wanted high-frequency detail | Lower Absorption, check Inverse, then lower final Reduction |
| Thin voice or missing bass note | Hum notch or high-pass filter is too wide or too high | Narrow the band, reduce the cut, or use a dedicated harmonic tracker |
| Hollow/phasey music | Too many harmonic notches or overprocessing in multiple stages | Remove unproven bands and rebuild from the fundamental |
| Pumping between phrases | Gate threshold/release or changing denoiser gain | Slow the transition, reduce range, or use manual clip gain |
Don’t judge a restoration at an exaggerated headphone level only. Check quietly, at a normal listening level, through speakers and headphones. A faint noise floor that disappears in context may be finished. An artifact that follows every syllable will remain distracting after export.
When Not to Remove All the Noise
Stop processing when the residual noise is less distracting than the side effects. Re-record when the performance can be repeated and the source problem is fixable. Preserve the original when the recording is archival, unique or historically important. For a badly contaminated one-off recording, create two deliverables if useful: a conservative archival master and a more aggressive listening copy.
A restoration tool cannot separate two signals that occupy the same time and frequency perfectly. Modern algorithms can make a better estimate, especially for speech, but the tradeoff remains. “Without quality loss” is not a setting; it is a listening judgment constrained by the source.
Remove Hiss and Hum FAQ
Where is DeHisser in Sound Forge?
In the current Sound Forge help, choose Effects → DeHisser. If it is missing, verify the installed edition and plug-in state instead of following a Sony- or MAGIX-era screenshot.
Should I remove hiss or hum first?
Remove stable hum first. Narrow notches or a dedicated de-hum processor can reduce specific tones before broadband hiss reduction analyzes the remaining floor. Reverse the order only when a controlled A/B on a copy clearly preserves more wanted audio.
Is electrical hum always exactly 50 or 60 Hz?
No. Those are common mains-related starting points, but the strongest recorded line may be shifted or appear at 100/120 Hz and higher harmonics. Measure the actual spectrum rather than entering a regional preset blindly.
How much hiss reduction should I use?
Use the least reduction that makes the hiss non-distracting. There is no safe universal dB value. Set the discrimination boundary while listening to Inverse or residue, then lower the final reduction until words, transients and fades sound natural.
Can a noise gate remove hiss?
A gate can lower hiss during pauses, but it does not remove hiss while speech or music is above the threshold. Use it after continuous-noise repair and keep the range and release natural enough that phrase edges do not pump. The Sound Forge noise-gate guide covers threshold, attack and release without pretending a gate is broadband audio noise reduction.
Why does cleaned audio sound metallic or underwater?
The processor is classifying part of the wanted signal as noise or chasing a changing noise floor. Improve the noise sample, reduce sensitivity and reduction, listen to the removed signal, or switch to manual automation. If the voice is barely louder than the noise, re-recording may be the only clean fix.
Can EQ remove 50/60 Hz hum?
Yes, when the hum is stable and only a few measured lines are objectionable. Use narrow peaking cuts at the actual fundamental and audible harmonics. A dedicated de-hum processor is safer when the ladder is long or the frequency drifts.
Can iZotope RX De-hum run in Sound Forge?
Yes, if the licensed RX plug-in is installed in a VST format that Sound Forge scans and supports. It is not automatically included with Sound Forge. If it does not appear, use the missing VST plug-in checklist.
The Practical Restoration Rule
Fix the source when you can. For an existing file, identify the noise, remove only the narrow hum lines you can prove are interference, and then reduce the remaining hiss conservatively. Listen to what the processor removes, protect the original, and stop when the recording sounds natural—not when the analyzer looks empty. That is slower than one-click suppression and much faster than undoing a brittle, hollow master.