I published an audiobook mastering suite that includes the correction tools that make a lot of these adjustments automatic.
One of these steps, as above, is RMS Normalize which sets the loudness to the middle of ACX specifications—assuming you change the setting a bit. Many of the tools in this suite depends on the tools that came before so be really careful if you decide to ad-lib.
You specify peak level “-3.4dB”, so that’s the limit to use …
The RMS value will go down a little after using the limiter,
so after limiting check the RMS value is still within specifications.
I used -3.5dB in the suite with the idea of leaving a little slop room for the conversion to MP3 for submission. The specification is peaks to be quieter than -3dB.
If you find yourself in Effect > Noise Reduction to make ACX Noise, try the correction suite on a raw chapter and see if it magically passes. Some microphones produce low frequency trash and rumble internally and it throws the other corrections off. That’s what Equalization > Low-Rolloff is there for.
You’ll probably want to use a compressor to boost the RMS, versus normalization which simply turns up the volume (which is why people are suggesting a limiter as well, so you don’t clip). Compression will keep you from clipping by reducing the volume of your peaks musically, versus just chopping them off with a limiter (which is basically a compressor with an infinite compression ratio).
Not all limiters are equal.
The Audacity limiter is deceptively sophisticated, and by default it provides “soft knee” limiting with “lookahead”, so it guarantees the maximum peak level, while compressing peaks quite gently (gently for a limiter). The “hard clipping” setting on the other hand, does “chop off” the peaks (brutally so). Limiter - Audacity Manual
After struggling with compressors and other audiobook processing for many, many forum posts, the mastering suite was found to hit the sweet (so to speak) spot of gently nudging your voice into compliance while:
a: Continuing to sound like you.
b: Producing corrections that ACX can’t hear.
ACX’s clear goal is listening fascinated to somebody over cups of tea while they tell you a story—not tell you a story over a bad cellphone. It’s certainly possible to beat your voice with a club and force it into Technical Compliance, but the second step in ACX acceptance is Human Quality Control. The humans do not like audible corrections.
Background noise is hard and that was left to the second part of the suite. There is no push the button and it all works (except in an April First posting I wrote). You can present with a very good, special purpose microphone in a bad room, or any microphone in a quiet, echo-free room.
I produced a passable voice clip with the laptop built-in microphones—in my quiet third bedroom.
It is remotely possible to have a celebrity voice that will not make it through the correction suite, but it hasn’t happened yet.