Archival audio recordings — whether they are digitized oral histories, vintage music tracks, radio broadcasts, or cassette tapes — often carry unwanted artifacts. Among the most common and distracting are hiss and hum. Hiss appears as a constant, high-frequency static, while hum introduces a low-frequency drone, typically at 50 Hz or 60 Hz depending on the power grid frequency of the original recording region. These noises not only degrade the listening experience but can mask important details in the original signal. Removing them effectively requires a methodical approach, combining the right tools with a solid understanding of audio restoration principles. This guide walks through the entire process, from diagnosing the noise to applying precise corrective techniques, ensuring you retain the integrity of the original material while achieving a clean, professional result.

Diagnosing the Noise: Hiss vs. Hum

Before applying any processing, it is critical to identify the type of noise present. Misdiagnosis can lead to poor results or even permanent damage to the audio. Use a spectral frequency analyzer (available in most audio editors) to visually inspect the recording.

Characteristics of Hiss

Hiss is broadband, meaning it spans a wide range of high frequencies, typically above 8 kHz. It is often white noise or pink noise shaped by the recording chain. Common sources include:

  • Analog tape hiss from magnetic particles
  • Electronic noise from microphone preamps
  • Amplifier noise in older equipment
  • Background air conditioning or environmental noise during recording

Hiss tends to be consistent throughout the recording, only varying in loudness if the original gain structure changed (e.g., during silent passages versus loud music).

Characteristics of Hum

Hum is tonal and narrowband — it concentrates at one or two specific frequencies. The fundamental is usually 50 Hz (Europe, Asia, Africa) or 60 Hz (North America, parts of South America, Japan), with harmonics at 100/120 Hz, 150/180 Hz, and so on. Hum sources include:

  • Ground loops between connected equipment
  • Electromagnetic interference from power cables
  • Poorly shielded audio cables
  • Faulty power supply filtering in vintage gear

Hum may change in level if the recording setup was not consistent, but its frequency remains stable, making it easier to target with notch filters.

Essential Tools for Audio Restoration

Professional results require capable software. While many free tools exist, investing in one or two high-quality restoration suites can dramatically improve outcomes. Below are the most widely used tools, each with strengths for hiss and hum removal.

Free and Open-Source Options

  • Audacity — A powerful, free audio editor with built-in Noise Reduction, Notch Filter, and Equalizer effects. Ideal for basic to intermediate restoration. Audacity official site
  • Ocenaudio — Intuitive interface with real-time preview and support for VST plugins. Good for quick edits.
  • Spek — A spectral analyzer (not an editor) useful for diagnosing hum frequencies.

Commercial Restoration Suites

  • iZotope RX — Industry standard for audio repair. Features like Spectral De-noise, De-hum, and De-clip are incredibly precise. Expensive but unparalleled. iZotope RX
  • Adobe Audition — Part of Creative Cloud. Offers Adaptive Noise Reduction, DeNoise, and manual spectral editing. Adobe Audition
  • Waves Restoration Bundle — Includes WNS (Waves Noise Suppressor), X-Hum, X-Click, and X-Crackle. Less expensive than iZotope.
  • CEDAR Audio — Used in broadcast and archival institutions. Very high-end, but most tools are only available as hardware or expensive plugins.

Preparing the Audio Files for Restoration

Proper preparation ensures the restoration algorithms have the best possible data to work with. Follow these steps before applying any processing.

1. Convert to High-Resolution Uncompressed Format

Work with a copy of the original file, never the original itself. Convert to WAV or FLAC (lossless) at the original sample rate and bit depth. Avoid working with MP3 or other lossy formats because the compression artifacts can confuse noise reduction algorithms. If the source is already lossy (e.g., an MP3 recording), consider it a last resort — results will be limited.

2. Normalize to a Safe Level

If the recording is very quiet or very loud, normalize the peak level to around -3 dBFS to leave headroom for processing. This prevents clipping during noise reduction. Do not apply dynamic compression before restoration; it can alter the noise profile and make removal harder.

3. Identify Clean Noise-Only Sections

Scan through the entire recording to find at least 0.5–2 seconds of pure background noise — no dialogue, music, or other desired content. This noise sample is essential for the "learn noise profile" method used by most tools. Silences at the beginning or end of a tape, or gaps between spoken words, are ideal.

Pro tip: If the recording has no quiet moments, consider recording a short section of the tape's leader or the ambient noise before the recording starts (if the analog source was transferred). Otherwise, you may need to use manual spectral editing, which is more advanced.

Removing Hiss: Step-by-Step Techniques

Three primary methods exist for hiss removal: spectral subtraction (noise reduction), spectral editing, and multiband expansion. The choice depends on the severity and complexity of the hiss.

Method 1: Spectral Subtraction (Noise Profile-Based)

This is the most common and effective approach for steady hiss. It works by analyzing the frequency spectrum of the noise-only sample and then subtracting that profile from the entire recording.

  1. Select the noise sample in your software (e.g., in Audacity, highlight 1–2 seconds of the noise-only section).
  2. Get the noise profile (Audacity: Effect > Noise Reduction > Get Noise Profile). In iZotope RX, use the Learn button in Spectral De-noise.
  3. Select the entire audio or the region you want to clean.
  4. Apply the reduction with careful settings:
    • Noise reduction (dB): Start with 12–18 dB. Too much will cause artifacts or remove harmonics.
    • Sensitivity: Controls how strictly the algorithm differentiates noise from signal. Higher values (e.g., 6–12 in Audacity) catch more noise but risk altering desired content. Lower values are safer.
    • Frequency smoothing (bands): Typically 3–6. More bands give finer control but can sound unnatural.
    • Attack/Release: In advanced tools like iZotope, set a slow attack (30–50 ms) to avoid "pumping" artifacts when the signal reappears.
  5. Preview and adjust. Listen carefully to a section with both foreground audio and hiss. If you hear warbling, "watery" artifacts, or loss of high-frequency detail, reduce the reduction amount or increase the smoothing.
  6. Apply in stages. For heavy hiss, it is better to run two passes of 50% reduction rather than one extreme pass. This gives more natural results.

Method 2: Spectral Editing (Manual)

When hiss is inconsistent or interwoven with the signal (e.g., a recording with constant wind or location noise), spectral editing provides surgical precision. Tools like iZotope RX's Spectral Repair or Adobe Audition's Spectral Frequency Display let you draw selections to remove or attenuate specific frequency regions.

  • Identify the hiss band. For example, tape hiss often dominates above 10 kHz.
  • Use a band-stop filter or EQ curve to gradually roll off frequencies above 10 kHz. A gentle slope of 6–12 dB per octave is less destructive than a brick-wall filter.
  • In spectral view, you can "paint out" short bursts of noise that occur only during quiet passages, such as a recorder motor hum that appears only during pauses in a voice recording.

Risk: Manual editing is time-consuming and can leave gaps or unnatural silence. Always listen to the result at normal volume to ensure you haven't removed essential sibilants or transients.

Method 3: Multiband Expansion (Dynamic Reduction)

This advanced technique is useful when hiss is only audible in quiet sections and becomes masked by louder content. A multiband expander (or downward expander) reduces gain in the high-frequency band when the signal level is low, leaving loud passages untouched.

  • Set a threshold just above the noise floor.
  • Apply a ratio of 1:2 to 1:4 — meaning when the signal drops below the threshold, the gain reduction is applied only to that band.
  • This preserves the natural dynamic range and avoids the "noise gate" click sound.

Most consumer tools do not include multiband expansion; it is available in iZotope RX's Spectral De-noise (in "Dynamic" mode) or dedicated plugins like Waves WNS.

Removing Hum: Targeted Techniques

Hum is easier to remove than hiss because it concentrates at known frequencies. However, careless removal can leave a "hole" in the low end or cause phase cancellation.

Using Notch Filters

The most straightforward method is a notch (band-stop) filter tuned to the hum fundamental and its harmonics. Most audio editors have a built-in Notch Filter effect.

  1. Identify the fundamental frequency using a spectrum analyzer. Run the recording through the analyzer and look for a sharp spike at 50 Hz or 60 Hz, and often at 100/120 Hz, 150/180 Hz, etc.
  2. Apply a notch filter at each affected frequency. Set the Q factor (bandwidth) very narrow — around 30–40 for aggressive removal, or 10–20 for a gentler cut that preserves adjacent material.
  3. Check for harmonics. Often only the fundamental and first harmonic are audible, but in extreme cases there may be several. Remove them one by one. Too many notches can cause ringing in the time domain.
  4. Listen for artifacts. A too-narrow notch can cause a "wobble" effect if the original recording had slight pitch instability (e.g., from tape wow). In such cases, use a broader notch (lower Q) and accept a slight loss of low frequencies.

Advanced De-hum Algorithms

Tools like iZotope RX's De-hum and Adobe Audition's Adaptive Noise Reduction can automatically track multiple hum harmonics and remove them with fewer artifacts. These algorithms use adaptive filtering that follows small frequency drifts (common in analog recordings).

  • Set the line frequency: choose 50 Hz or 60 Hz, and the tool will calculate harmonics.
  • Adjust the "number of harmonics" slider — start with 3–5. Too many can cause "electronic" artifacts in the midrange.
  • Use the "filter length" or "attack" control to balance removal vs. transient preservation. Longer filter lengths smooth the removal but may smear percussive sounds.

Manual EQ Curve (High-Pass Filter)

If hum is only the fundamental (e.g., 60 Hz) and the recording contains no useful audio below 80–100 Hz, you can apply a high-pass filter with a steep slope (24 dB/octave) set just above the hum frequency. This eliminates both the hum and any subsonic rumble. However, be careful: many archival recordings have important low-frequency content (bass guitar, pipe organ, footsteps, etc.). Use this method only when the low end is not essential.

Warning: Do not use a high-pass filter if the recording contains dialogue with deep male voices, as you will remove fundamental frequencies of the voice (typically 80–180 Hz). The voice will sound thin and "telephone-like."

Combining Techniques for Complex Noise

Many archival recordings contain both hiss and hum, plus additional imperfections like clicks, pops, crackle, and even clipping from analog overload. A workflow that addresses each noise type in the correct order yields the best results.

  1. Click and crackle removal first — Use a declicker tool (e.g., iZotope RX De-click, Audacity's Click Removal) before any broadband processing. Clicks can interfere with noise profile learning and can be amplified by later EQ.
  2. Hum removal second — Because hum is tonal, removing it before hiss reduction ensures the hum does not contaminate the noise profile. Also, if you use a high-pass filter, do it now.
  3. Hiss reduction third — Apply spectral subtraction or spectral editing after hum is gone. The noise profile will be cleaner, and the algorithm will not try to "subtract" hum energy.
  4. Final equalization and dynamics — After all noise removal, the audio may sound dull or unnatural. Use a gentle EQ boost in the high frequencies (e.g., shelf filter at 8 kHz up 3 dB) to restore presence. If needed, apply a compressor or limiter to even out levels.

Real-World Example: Restoring a 1960s Oral History Tape

Let's walk through a typical restoration using free tools (Audacity) and illustrate the decision-making process.

Step 1: Prepare and Analyze

  • Open the WAV file (44.1 kHz, 16-bit) in Audacity.
  • Switch to spectrogram view (View > Spectrogram). You see a thick band of noise above 8 kHz (hiss) and a persistent line at 60 Hz with a faint line at 120 Hz (hum).

Step 2: Remove Hum

  • Use Effect > Notch Filter. Set Frequency: 60 Hz, Q: 30. Apply.
  • Repeat for 120 Hz with the same Q. After the second notch, the hum is gone. Check the spectrogram — no visible line at 60 or 120 Hz anymore.

Step 3: Remove Hiss

  • Find a 1.5-second gap between spoken sentences (only hiss). Select it.
  • Go to Effect > Noise Reduction > Get Noise Profile.
  • Select the entire track. Open Noise Reduction again. Settings: Noise reduction (dB): 15, Sensitivity: 6.00, Frequency smoothing (bands): 3.
  • Preview. The hiss reduces but the voice sounds slightly "underwater." Reduce Noise reduction to 12 dB and decrease Sensitivity to 4.00. Preview again — better balance.
  • Apply the effect. The hiss is now 80% gone, but a faint amount remains.
  • Run a second pass with a different noise sample from a later quiet section (the hiss might be slightly different due to tape wear). Use the same settings but reduce Noise reduction to 8 dB. This final pass removes most of the residual hiss without further distortion.

Step 4: Polish

  • Apply a gentle high-shelf EQ boost of +2 dB at 6 kHz to restore brightness lost during hiss reduction.
  • Normalize to -1 dB peak.
  • Export as 48 kHz, 24-bit WAV for archival.

Common Mistakes and How to Avoid Them

Even with the best tools, mistakes can ruin a restoration. Watch out for these pitfalls.

Over-Reducing Noise

Pushing noise reduction too far creates "musical noise" — warbling, metallic artifacts that are more annoying than the original hiss. Always listen in context. It is better to leave a small amount of hiss than to destroy the audio's natural timbre.

Removing the Noise Along with the Signal

Noise reduction algorithms operate by comparing the noise profile to the signal. If the foreground audio has frequency content that overlaps with the noise (e.g., a woman's voice sibilance matches the hiss frequency band), the algorithm will attenuate it. To minimize this, use a wider frequency smoothing and lower reduction amount. Or, use spectral editing to manually protect the sibilant sections.

Applying Noise Reduction to the Entire File Without Checking

Different sections of a recording may have different noise characteristics (e.g., the beginning of a tape may have less hiss than the end due to tape wear). Apply noise reduction only to sections that need it. You can use the "Selection" tool to isolate problematic regions.

Skipping the Backup

Always work on a copy. If you apply destructive processing (changing the file directly), you cannot undo. Use non-destructive workflows (e.g., Audacity's "destructive" effects — save a copy before) or work with a project that stores undo history. Better yet, always keep the original unprocessed file.

External Resources and Further Reading

Conclusion

Removing hiss and hum from archival audio is a delicate art that requires patience, careful listening, and methodical application of tools. By understanding the nature of the noise, preparing the file correctly, and applying a staged workflow — remove hum first, then hiss with spectral subtraction or spectral editing — you can clean up even heavily degraded recordings while preserving their historical and emotional integrity. Always remember: the goal is not total silence, but a natural sound that honors the original performance or speech. With practice, these techniques become second nature, and your restored audio will stand as a testament to the power of modern audio restoration.