The proposed experiment, which appears to be to simply try to create a low-level signal and try to hear it, lacks refinements for which I will try to make a logical care for.
The first thing is that in all reasonable likelihood you really don't want to try to listen for the lowest level recording that you can hear.
Rather you want to listen for the lowest level sound that you can hear played in an audio system with its gain set at a reasonable, comfortable and real-world level for the recording at hand. So it might be more likely that you want to listen to a recording of a sound presented at a reference level that is comfortable for you and then you want to listen to the low-level test sound(s).
Your goal is to set the reference sound for a reasonable, comfortable and real-world SPL such as the one you would set your system for were you going to do some engaged listening.
One way to accomplish this is to listen to the reference sound and adjust it until it sounds right to you, given that your goal is engaged comfortable listening.
Secondly, you don't want to go straight from listening at a reference level straight to the lowest level, because that is typically not what happens in real-world listening. In most music and dialog the periods of high loudness and low loudness are separated by transitional periods.
Thirdly you don't want to plunge the listener cold turkey into the most critical listening situation. Again, it is easier for people to do well in listening tests if they transition from easy even non-challenging listening to challenging listening in a gentle progression.
Fourthly, you don't want to create a crisis where the listener has to do or die. Instead, you want to create a situation where less-than-perfect performance is rewarded a little while preserving a good reward for better performance.
Finally, you want an experiment that is open-ended, because we are dealing with people, and people always have the potential to surprise y0u with how well or how badly they perform. We want answers from the real world, not answers of our choice or based on our foregone conclusions.
With all those things in mind. I humbly present for your entertainment an ergonomically designed listening test, encapsulated in a file that you can download from here:
https://www.dropbox.com/s/n50g7b6hs14x8dj/high resolution listening test audacity HD.flac?dl=0
The instructions are for you to adjust the first two repetitions of the recorded vocal sound for comfortable, engaged listening, and report back the number of times you can hear the sound repeated, including the first two repetitions.
Please note that the actual detailed characteristics of the sound are less important since the goal of the evaluation is to hear how much the same identical sound can be attenuated and still be heard. If it were in the realm of the peak sensitivity of the ear, then it would be set lower during the initial reference level setting, If the sound was in some realm where the sensitivity of the ear was less, then following the instructions would result in it being set up with more gain and actual SPL level