Why Your Podcast Sounds Noisy (and How to Fix Each Cause)

2026-10-16 · Formavide

You hit record, talked, played it back — and there it is: a wash of hum, hiss or echo under your voice. Noise in podcast audio always has a specific cause, and each cause has a specific fix. Here's how to diagnose yours and what to do about it.

AI denoising with an adjustable strength dial and A/B preview. Runs in your browser, nothing uploads. Free to try.

First, identify which noise you have

"Noisy" is four different problems wearing one label. Play a silent section of your recording with eyes closed — each noise has a fingerprint:

The causes worth fixing at the source

Post-processing can rescue a take, but three culprits are far easier to prevent than to repair:

• Room echo. Bare walls are the enemy. The cheapest fix in audio: record under a thick blanket, in a closet full of clothes, or facing an open bookshelf. No plugin fully un-echos a voice once reflections are baked in.
• Mic distance. Every doubling of distance costs you ~6 dB of voice — which you'll compensate with gain, which raises the noise floor with it. A fist's width off-axis from the mouth (slightly to the side, dodging plosives) beats any de-noiser.
• Ground loops and buzz. If the hum changes when you touch the mic or laptop charger, it's electrical — try a different outlet, unplug the charger, or use the second USB port before you reach for software.

The post-production rescue pipeline

For the noise that made it into the take, order matters. Each step should happen once, in this sequence:

1. Trim the dead air. Cut the silent head and tail first — there's no point denoising minutes of room tone nobody needs. A local trimmer does this in seconds without uploading the episode.
2. Denoise with a dial, not a button. AI denoisers work best with discretion: start around 60–70% strength for steady hum and hiss, and listen to a segment before and after at each setting. Full-power denoising on a light-hum recording is how voices start sounding like they're under water.
3. Normalize the loudness. Denoising often reveals how quiet the take was. Bring it to a consistent level — but use a tool that leaves silence alone and caps the boost, or you'll amplify the noise floor right back into audible range.
4. Then, and only then, compress. A compressor after cleanup evens out the performance. Compress before denoising and you've raised the noise into the compressor's grasp.

Why "AI denoise" beats the old manual chain

The traditional recipe — noise gate, notch filter, spectral repair — assumes you'll spend an hour learning parameters per episode. Modern speech-enhancement networks (like DeepFilterNet3) learned what clean voice sounds like from thousands of hours of examples, and simply remove everything that isn't that. For steady noise the result is a one-pass fix that used to take an afternoon.

Two honest limits: it's voice-centric, so music under the voice gets treated as noise (that's a job for a vocal splitter, not a denoiser); and it can't recover a voice that was drowned completely — it makes marginal takes usable, not studio-grade.

A note on where you run the fix

Episode audio is personal: early drafts, unflattering outtakes, guest recordings under NDA. A browser-based denoiser keeps the file on your machine — you can watch the network tab stay silent. The tool we build runs the AI model in your tab with an A/B preview, so you can judge the strength setting by ear before committing the whole episode. Free to try, and the same pipeline works on video files without touching the picture.