My project is to build a small sonometer (SPL, VU meter, sound meter, decibelmeter) with a microphone, a microcontroller and a LED bar. There are numerous projects already avialable. Enough to build my own Exemple 1 Exemple 2
The real challenge is here: I would like the LED bar to react only (or mostly) to human voice (or sounds, spoken or sung) independently to the daily environment, more or less noisy. Should I use a physical filter on the microphone and/or could I use a filter to recognize a voice/sound from the noise?
Any suggestion is welcome. Thank you for your help.
I don't think that is currently possible with any Arduino or other MCU.
However, I have seen a demonstration of the idea by researchers from Google, who used a machine learning algorithm running on an extremely powerful computer to extract the voice component from the background noise.
Unfortunately, to subtract noise from voice+noise requires that you understand the noise rather well.
The free program Audacity (for PCs) does pretty well, if you give it a good recording of the ambient noise, taken with the same equipment and settings as the voice+noise signal, as a starting point.
Normally you mean a device to attenuate (not remove) a range of frequencies. By definition noise is random so it will escape attenuation when it is outside the range you want to attenuate.
So what do you want to adjust with this mythical filter?
What ever it is it will not do what you want because you don't seem to know the theory of how sound works.