How to run a blind A/B test on your mix at matched loudness
When you know which version you are hearing, that knowledge comes along. The version you just spent an hour on, the expensive compressor, the outboard box on the rack: it is easy to hear what you expect. A blind test takes the names away, so the only thing left to judge is the sound.
Two conditions make a blind test fair. Both versions play at the same loudness, because the louder one tends to win. And you switch while the song keeps playing, so you compare the same passage.
What you need
- A DAW that runs AU, VST3 or AAX plug-ins (see the DAW list in the README)
- Kirin Hypha, a free, open-source plug-in for macOS and Windows
Before and after your processing
Insert Hypha's PRE before the plug-ins you want to judge and POST after them. On POST, press BLIND. Hypha fixes a matched level and hides the names. Switch between SOURCE 1 and SOURCE 2 as often as you like without stopping playback. Once both have sounded, REVEAL shows which was PRE and which was POST.
The full setup is in How to compare before and after your processing at the same loudness.
Rounds: is it a preference or a coin flip?
One pick can be luck. Set ROUNDS to 5 or 10 on the BLIND screen. In each round you pick the one you prefer, and the positions are shuffled between rounds. The result shows how many times you chose each side. When one side leads by two or more, it also shows how often a result at least this uneven would occur at random.
SAVE IMAGE on the result screen saves a 1080 × 1080 PNG with the round count and the result, if you want to share it. Hypha does not store your answer, and the audio is untouched. It is not an ABX test: it tells you which one you prefer, not whether you can tell them apart.
Two processors: outboard against a plug-in
Put PRE1, A, PRE2, B and POST in that order on one track. Put outboard gear on a hardware insert in A and a plug-in in B (two plug-ins work the same way). On POST, press A/B and choose PRE2. During A/B, PRE2 sends the original input to B, so A and B process the same source. LISTEN, MATCH and BLIND then compare A and B at the same loudness. With ROUNDS set to 5 or 10, A/B asks which one you prefer before REVEAL, so you make your call before you see the answer.