How to Benchmark Audio Tools at Home With Nothing but a Phone and a Fan

You can run a fair comparison of noise suppression tools in an evening, using a desk fan, a phone and the computer you already own. The trick is not fancy equipment. It is controlling conditions so that the only thing changing between recordings is the software. Do that, and your ears are a perfectly adequate measuring instrument for the question that matters: which tool makes me sound best, in my room, with my noise?

Why a home test beats someone else’s verdict

Published results depend on the reviewer’s microphone, room and choice of noise. Yours will differ. A tool that handles an air conditioner beautifully might stumble on the specific rattle of your radiator or the television next door. Since many products in this category offer a trial or a free tier, and the suppression in meeting apps costs nothing, you lose only time by checking for yourself.

What you need is modest. A fan gives you steady broadband noise at two or three repeatable speeds. A phone gives you unsteady noise on demand: a podcast for competing speech, a video of a barking dog, a recording of keyboard clatter. Add a tape measure, some masking tape and a short script to read.

Setting up so the comparison is fair

Fix the geometry

Put the microphone where you normally use it and mark its position with tape. Mark where your chair goes. Place the fan a measured distance away, off to one side, pointing away from the mic so air does not blow across it, since moving air creates rumble that has nothing to do with the test. Set the phone at a second marked spot with its volume at a level you note down. Every recording must use the same positions and the same fan speed.

Fix the script

Write five or six sentences and read the same ones every time, at the same pace and loudness. Include quiet consonants, since those are what aggressive suppression eats first: words with “f”, “th”, “s” and “h”, plus a few numbers and names. End with ten seconds of silence so you can hear what the tool does when nobody is talking.

Fix everything else

Turn off every other layer of processing. If you are testing a standalone utility, disable the suppression in the recording app and any enhancement in the operating system, otherwise you are grading a stack and not a product. Use the same recorder throughout, and do not touch the input gain once it is set.

The recordings to make

Start with a clean baseline: no fan, no phone, no suppression. This is your reference for how your voice should sound. Next record the noisy baseline: fan on, suppression off. Then, for each tool and each strength setting you care about, record the script against the fan alone, against the phone playing speech, and against the phone playing sudden sounds. Put the tool, setting and noise type in each filename, because after the sixth take you will not remember which was which.

If you want to see how a dedicated benchmarking site structures the same idea, with one method applied to every product, browse the published tests at SignalBench and borrow whatever parts of the approach suit your setup. A home version will never be as controlled, and it does not have to be. It only has to be consistent with itself.

Listening without fooling yourself

Wear headphones, match the playback volume between clips, and listen to each file for two separate things. First, noise: how much is left during speech, not only in the gaps, and does it pump in and out between words? Second, voice: compare it with your clean baseline. Are word endings intact? Did the quiet consonants survive? Is there a warble or a hollow, underwater color? Does the first syllable after a pause get clipped?

Bias is the real enemy here. If you know which clip came from the tool you paid for, you will hear it as better. Ask someone to rename the files with random letters and keep the key, or write a note of the mapping, shuffle the copies and come back the next day. Score each clip from one to five for noise and separately for voice before looking anything up.

Be clear about what this test cannot show. It will not reveal behavior on a poor network, performance with other voices and accents, or stability across a three-hour call. It says a great deal about your desk and little about anyone else’s.

What to do with your scores

Pick the tool whose voice score stays close to your clean baseline at the lowest strength that makes the noise tolerable. That combination, natural speech first and sufficient quiet second, is what listeners experience as good audio. Save the baseline files too. The next time you change a microphone, a room or an app, you can rerun the same script in fifteen minutes and know whether things got better or merely different.