Hum and Buzz in a Recording: Tracking Down Mains Noise and Cleaning It Up

Hum, buzz and whine: name what you hear

Electrical noise comes in a few recognizable flavors, and the name points to the cause.

Hum

A low, steady, smooth tone, felt as much as heard, at the mains frequency

Buzz

A harsher, raspier version of hum, rich in harmonics that reach into the voice range

Whine

A higher tone that rises and falls with computer activity, screen changes or charging

Rhythmic chatter

A pulsing dit-dit-dit pattern, typical of a mobile phone close to a microphone or cable

Broadband hiss, traffic and air conditioning are different problems with different tools, covered in speech recognition and background noise.

Where electrical noise gets into a recording

Alternating current reverses direction 50 times a second in Europe, most of Africa, Asia and Australia, and 60 times a second across North America and some other countries; Japan uses both, depending on region. Equipment connected to it can leak that frequency, and multiples of it, into audio. The usual routes:

• Ground loops. When two pieces of equipment are each grounded through their own power outlets and also connected by an audio cable, a small current can flow along the cable shield and appear as hum.
• Unbalanced cables near power cables. Guitar-style jack cables and consumer phono cables pick up fields from power leads they run beside.
• USB power and chargers. A laptop on its charger, connected to a USB microphone or interface, can inject hum or whine; the same setup on battery is often silent.
• Dimmers, LED drivers and fluorescent fixtures, which switch current sharply and radiate buzz.
• Phones and wireless devices next to microphones, cables or interfaces.
• Faulty or poorly shielded cables and connectors, where hum comes and goes when the cable moves.

Identify the source by ear and by spectrum

1. Find a pause in the recording and listen on headphones. Decide whether you hear hum, buzz, whine or chatter.
2. Open the file in an editor with spectrum analysis, such as Audacity's Plot Spectrum, and select a few seconds of pause. Sharp peaks at 50, 100, 150 Hz or at 60, 120, 180 Hz confirm mains hum. Peaks that do not sit on those multiples suggest another source.
3. Before the next recording, monitor the input live on headphones and remove suspects one at a time: unplug the laptop charger, move phones away, switch off dimmers and nearby lights, unplug unused cables.
4. Power all connected audio equipment from the same outlet or power strip, which often breaks a ground loop.
5. If hum changes when you touch or move a cable, replace the cable.

Never remove or defeat the safety ground on a power plug to stop hum. If a ground loop persists, use equipment designed to break it safely, such as an audio isolation transformer or a DI box with a ground lift switch.

Filters that remove hum, and how far to go

Steady mains hum is one of the few noises that filtering removes well, because it sits at known, narrow frequencies. Three tools cover most cases:

• A high-pass filter around 70 to 80 Hz reduces the fundamental and low rumble with little effect on most voices.
• Notch filters remove very narrow bands at the fundamental and each harmonic. Narrow notches take away little of the voice.
• A de-hum effect, found in many audio and video editors, places notches at the fundamental and harmonics for you; some track small changes in the hum's frequency.

With ffmpeg, a set of narrow band-reject filters does the same job. For 60 Hz hum:

ffmpeg -i interview.wav -af "bandreject=f=60:t=q:w=30,bandreject=f=120:t=q:w=30,bandreject=f=180:t=q:w=30,bandreject=f=240:t=q:w=30" interview-dehum.wav

For 50 Hz hum, use 50, 100, 150 and 200 instead. Start with the fundamental and a few harmonics, listen, and add more only if the hum is still audible. Buzz with a broad spread of harmonics may need a noise-profile reduction tool on top, applied gently.

A worked example: a podcast with a charger hum

Does hum affect transcription?

Steady low hum below the main voice range often has limited effect on speech recognition, since models are trained on many imperfect recordings. Strong buzz is a different matter: its harmonics extend into the frequencies that carry consonants and can mask quiet words. Hum also makes transcripts harder to check by ear and tiring to listen to. Clean it before transcribing when you can hear it clearly over the speech, and judge the result with your ears rather than by how flat the spectrum looks.

Limits and trade-offs of hum removal

• Every notch removes a sliver of the voice too. Many wide notches make speech sound hollow or phasey.
• Hum that drifts in pitch, as from some generators, is harder to track with fixed notches.
• Buzz from dimmers and switching power supplies has broadband components that notches miss.
• Hum louder than the voice cannot be removed without damaging the speech.
• Removing hum from a clipped or heavily compressed file will not fix the distortion; see clipped audio for that problem.
• Filtering every file with the same preset can remove hum that is not there, along with some voice.

Preventing hum before the next recording

• Record laptops on battery when using USB microphones or interfaces, or test both ways.
• Plug all connected audio gear into one outlet or power strip.
• Use balanced XLR cables for microphones, and keep audio cables away from power leads, crossing them at right angles where they must meet.
• Keep phones in airplane mode and away from microphones and interfaces.
• Avoid recording under dimmed lights; switch them fully on or off.
• Record a 30-second test and check it on headphones before every session; checking audio quality has a quick routine.

What mydubly does and does not do with hum

mydubly does not remove hum or noise. The browser decodes the file's audio locally to 16 kHz mono, splits it into chunks of about 30 seconds and sends compressed chunks over HTTPS for transcription with Whisper. So clean the audio in an editor first, export it as WAV or FLAC, and upload the cleaned file. mydubly accepts MP3, WAV, M4A, AAC, OGG and FLAC audio, and MP4, MOV, WebM, MKV and M4V video, up to 2 hours per file.

You get a plain and timestamped transcript, SRT and VTT subtitles and, optionally, a translated transcript. Keep the page open until the job finishes, since leaving or refreshing cancels it; credits reserved for a job that ends without a result are returned.

When another tool is the better choice

For severe buzz, hum mixed with other noise, or valuable archive recordings, dedicated audio repair software or a professional restoration engineer will get further than basic filters. For recurring problems in a fixed studio, an electrician or audio installer can trace ground issues safely. For podcasts that are otherwise clean, notch filters and better cable habits are usually enough; the podcast transcription page covers what comes after.

Hum can also lead recognition to produce stray phrases in otherwise silent passages; Whisper hallucinations describes what to look for.

Next step: clean a test excerpt

Filter a one-minute excerpt with notches at your mains frequency and its harmonics, compare it with the original on headphones, and transcribe the cleaned version with audio to text. Transcripts cost 1 credit per minute with a 5-credit minimum, so an hour-long episode costs 60 credits (6¢). Video recordings can go through video to text.