From: DavMac AT iname DOT com (Davin McCall)
Newsgroups: comp.os.msdos.djgpp
Subject: Re: 8-Bit Clipping
Date: Wed, 18 Aug 1999 03:19:43 GMT
Organization: Monash Uni
Lines: 88
Distribution: world
Message-ID: <37ba254f.9723519@newsserver.cc.monash.edu.au>
References: <37B46FE9 DOT 87C3B20D AT geocities DOT com> <1097 DOT 990817 AT Phreaker DOT net>
NNTP-Posting-Host: damcc5.halls.monash.edu.au
X-Trace: towncrier.cc.monash.edu.au 934946329 152 130.194.198.138 (18 Aug 1999 03:18:49 GMT)
X-Complaints-To: abuse AT monash DOT edu DOT au
NNTP-Posting-Date: 18 Aug 1999 03:18:49 GMT
X-Newsreader: Forte Free Agent 1.1/32.230
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com

Adding the samples is the only way to properly mix two sounds. It is
fine to scale each sample stream, but each of the various streams will
then have a reduced volume.

In terms of a real-time mixing system with multiple channels (probably
fixed number, but not necessarily) this wouldn't be so bad, as a sound
by itself should be the same volume as when it was played together
with other sounds - just so long as a constant scaling rate was used.

If, however, when one sound is playing it is played at full volume,
yet when two are playing they are scaled by a factor of 50%, then the
two will sound noticably quieter.

Davin.


On Tue, 17 Aug 1999 02:19:54 +0700, Batchex <thedark1 AT Phreaker DOT net>
wrote:

[...]
>produce a "wick" or a "whack". For that reason, I listed a few ways to
>mix sounds and prevent clipping (theoretically, note that I haven't
>got the time to test these approaches yet, so tell me if it works).
>
>Assuming 8 bit, signed data format. Length is the length of the
>longest sound file that is to be mixed.
>
>1. Scale each samples with a certain values before adding them
>together. The sum of all scale values must be 1.
>ex.1 : two channels (sound files), each channels is weighted 0.5 (the
>sum of all weights is 0.5 + 0.5 =1).
>
>    for(i=0;i<length;i++)
>        mixed[i] = (channel1[i]*0.5) + (channel2[i]*0.5) + 0.5;
>
>
>2. Add the samples from each channels together, scan the min and max
>values of the resulting buffer, then scale the all the samples
>resulted from the mixing with the ratio of the max (or abs(min), which
>ever the greater) of the scanned sample values and the max value the
>data format supports.
>ex.2: data format is 8 bit signed, so the min value the data format
>supports is -128, and the max is 127.
>
>         int buffer[length];
>         int sampleMax, sampleMin;
>         float scale;
>
>         sampleMin = 2147483647;
>         sampleMax = -2147483678;
>
>         for(i=0;i<length;i++)
>         {
>          buffer[i] = channel1[i] + channel2[i];
>          if(buffer[i] < sampleMin) sampleMin = buffer[i];
>          if(buffer[i] > sampleMax) sampleMax = buffer[i];
>         }
>
>         if((sampleMax+sampleMin) >= 0) scale = sampleMax / 127;
>         else scale = sampleMin / 128;
>
>         for(i=0;i<length;i++)
>             buffer[i] = (int)((buffer[i] * scale) + .5);
>
>
>3. The approach no 2 can be simplified if there is a fixed number of
>channels to be mixed. The result of the addition of each samples from
>each channels is scaled by 1/number of channels.
>ex.3: number of channels is 4, so the scale value is 1/4 or 0.25.
>
>      for(i=0;i<length;i++)
>      {
>          buffer[i] = channel1[i] + channel2[i] + channel3[i] +
>                      channel4[i];
>          buffer[i] = (int)((buffer[i] * 0.25) + 0.5);
>      }
>
>
>Hope that helps.
>
>Batchex
>mailto:thedark1 AT Phreaker DOT net
>
>

__________________________________________________________
       *** davmac - sharkin'!! davmac AT iname DOT com ***
my programming page: http://yoyo.cc.monash.edu.au/~davmac/