How may I normalize a (voice) audio mp3 or aac file with no loss, having the gain rised as much as possible (mitigating distortion using a compressor), so in a long conversation people that speak softer can have more gain for their voice and people that speak louder can have less gain?
Normalization will only move the audio up to the point where there is no distortion.You will probably want to "squash" the audio using dynamic audio compression and then increase the gain. Compression isn't really lossless in the sense that you are changing the audio and ultimately the dynamics.
A compressor is a tool that engages at a certain audio level threshold. When the audio exceeds the threshold (for example -12dB), it lowers the levels of the louder sounds by the ratio (for example 2:1). 2:1 is saying that if the audio is 2dB over the threshold, then the output will be 1dB over the threshold. You usually can express attack and release values which specify how quickly the level of audio changes over the specified time value.
Just to note the compression I described is downward compression. There are other types as well.
The overall audio level is reduced since peaks are pushed downward, so you can push the audio gain upwards to produce a louder sound. A compressor is a valuable, but easily abused tool (think pop music released in the last 10 years...). You have to do much experimentation to get it to sound right.
It may also be a valid option to just boost volume in the specific quieter sections (or lower the louder sections) by drawing volume envelopes. Since this is just voice audio, I'm not sure if there is a specific "need" to want to increase the overall volume, but rather just to have the audio be somewhat consistent. If you have one person that is capable of mumbling at the start of a word and shouting by the end of it, compression is probably a better choice.
I'm not sure what tools/OS you have available to you, so this answer is tool agnostic.