Re: [SoX-users] Fwd: Using sox to normalize rms and peak

[Sergei S. <ser...@ya...>]

----- Original Message -----
> From: Fmiser <fm...@gm...>
> To: sox...@li...
> Cc: 
> Sent: Thursday, December 13, 2012 3:35 AM
> Subject: Re: [SoX-users] Fwd: Using sox to normalize rms and peak
> 
>>  Francesco Danieli wrote:
> 
>>  Hi we are trying to use sox to normalize a bunch of audio
>>  files to a constant peak and RMS level.
>> 
>>  Ideally sox filename -e stats should read
>> 
>>  Pk lev dB Overall 0
>>  RMS lev dB Overall -9
>> 
>>  for all our files.
>> 
>>  We managed to obtain the first value with -n option.
>>  What we are missing is how to get RMS level of the files to
>>  the desired db,leaving Pk level untouched.
>> 
>>  We tried "gain -ne"  and "gain -nb" but they both vary 
> the Pk
>>  level. We also tried loudness option but it does not give us
>>  what we are looking for.
> 
> The difference between RMS and peak level is called "crest
> factor".  It cannot be change by changing the gain.  The only
> way to change the crest factor is with a limiter, compressor,
> expander, or gate.  These tools can change the dynamic range and
> thus the crest factor. 
> 
> But what you are asking is an odd request - so I'm now really
> curious _why_ you want normalize both peak and RMS.
> 
> --    Philip
> 

"The only way to change the crest factor is with a limiter, compressor, expander, or gate." - not exactly.

All-pass filter can change it too. But I am not aware a bullet-proof recipe. I.e. the filter, depending on what it is and on input signal is can both increase and a decrease crest factor.

Tu understand my statement try to visualize a signal consisting of, say, fundamental and second harmonic:

f(t) = A1 * sin(w * t) + A2 * sin(2 * w * t + phi)
.

If phi = 0, then peak amplitude will be A1 + A2, i.e. peaks of the two harmonics with coincide with the 1/W period.

Making phi non zero can decrease the peak amplitude because for, say, phi = pi/4 the peaks never coincide.

Regards.
  Sergei.