You are right if you are talking of mixing lots of tracks. But for one
time normalization of a final track (that what we are talking about) it is
absoulutely irrelevant, where my samples are from (single sample from
analog or mixed from 256 sources). I have 16 Bit Samples and I introduce
an maximum error of +-1! No quality loss at all.
This would be different if I do lots of subsequent calculations in
integer, especialy when I lower the amplitude, storing then result in
integers.
DAVID BALAZIC <[EMAIL PROTECTED]> on 17.01.2000 17:03:43
Please respond to [EMAIL PROTECTED]
To: [EMAIL PROTECTED]
cc: (bcc: Thomas Marschall/MUC/Santix)
Subject: Re: [MP3 ENCODER] Normalization routine?
Date: Mon, 17 Jan 2000 15:47:52 +0100
From: [EMAIL PROTECTED]
>I think performance loss due to rounding errors can be ignored because:
You do mean _quality_ loss ?
>rounding errors are _very_ small
> (if we take 16 it samples we have values from -32767 to +32768
> if we assume we have a sample around +-16000 and double
> the amplitude we get around +-32000 with an error of +-1
> this is less than 0.0033% (damn no calculator am I right?))
>
>show me a AD device that samples this accurate. The error introduced
>by simply multiplying a constant value is (IMHO) minimal.
But some audio records ( like the music CDs ) are not simply sampled,
but mixed from a lot of tracks. When you mix two tracks you need one extra
bit to avoid resolution loss. So if you sample with 8 bit and then you
mix 256 channels ( in practice not that many is used , I know ) then you need
8 extra bits ( giving a 16 bit result ) !
Good audio software uses dithering when converting the results of a
computation from some more accurate format to 16 bit to achieve better
results. ( now go to them and say they are wasting their time; 16 bit
is anough for everyone :-)
OTOH , if the samples are converted to some more precise format in LAME,
then the normalization should be done after the conversion, because it
does not make it slower and it does have higher quality , and quality
is the point of the LAME project , no ?
>DAVID BALAZIC <[EMAIL PROTECTED]> on 17.01.2000 14:41:05
>
>Please respond to [EMAIL PROTECTED]
>
>
>
>
>
>
>
>
> To: [EMAIL PROTECTED]
>
> cc: (bcc: Thomas Marschall/MUC/Santix)
>
>
>
> Subject: Re: [MP3 ENCODER] Normalization routine?
>
>
>
>
>
>
>
>
>From: Ross Levis <[EMAIL PROTECTED]>
>
>>I can't help you directly but I always normalise before encoding music for my
>FM radio
>>station. I use the CD rippers Audiograbber and CDex which both have this
>option.
>>CDex being freeware, you may be able to obtain code from the author
>>(mailto:[EMAIL PROTECTED] http://www.cdex.n3.net)
>>
>>Ross.
>
>By normalizing 16 bit samples you loose quality,
>due to rounding errors.
>But maybe the loss can be reduced by doing the normalization
>on when the samples are in more suitable format ( double or something ) ?
>
>
>--
>David Balazic , student
>E-mail : [EMAIL PROTECTED] | living in sLOVEnija
>home page: http://surf.to/stein
>Computer: Amiga 1200 + Quantum LPS-340AT
>--
>--
>MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
>
>
>
>--
>MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
--
David Balazic , student
E-mail : [EMAIL PROTECTED] | living in sLOVEnija
home page: http://surf.to/stein
Computer: Amiga 1200 + Quantum LPS-340AT
--
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
