What is the ideal decibel? Do we need to adjust volume of datasets?

Deep Speech
#1

Hi, I’m training speech recognition model for Japanese with my own datasets.

Parts of my datasets are very low audio.
Should I increase volume of such audios before training?
If I normalize audio, is it better?

What average decibel is ideal?
I mean I may have to decrease volume when too loud.

Thanks in advance.

#2

When used in production do you expect the volume of the audio to change significantly or do you expect all the audio to be approximately of the same volume?

#4

Volume can be approximately the same even though i’m not sure i can release it as a real product…

Thanks for your advice.