Re: [Iaxclient-devel] sound issues

[Eloy D. <elo...@vo...>]

Hi Steve,


Thanks for the feedback! I just discovered that Borland C++ and gcc behave
differently. I can do a log10(0) in gcc without causing runtime errors, but
with Borland C++ it generates a "sing error" but still returns "-inf" for
log10(0). It seems to be warning message. The irritating thing is that the
message pops up in a message box :-S

Hopefully I can disable this 'feature'...


Regards,
Eloy


On Mon, 26 Apr 2004 16:31:32 -0400, Steve Kann wrote
> Eloy Domingos wrote:
> 
> >Hi Steve,
> >
> >
> >Completely correct, log10(0) is -infinity :-). This will also give you
> >a runtime singularity error and break your program, so this exception
> >needs to be handled properly, as a well-behaved program should... :-p
> >  
> >
> I suppose this depends on what you're trying to do with the value 
> later.  You don't get any error, for example, if you try to do this:
> 
> #include<math.h>
> 
> main() {
>   printf("%f\n", log10(0));
> }
> 
> Which happily works for me:
> 
> stevek@canarsie:~ $ /tmp/a.out
> -inf
> 
> Most clients will use this value to show some kind of volume level 
> meter.  Depending on how you implement your meters, you may or may 
> not have any trouble with -infinity..
> 
> For example, in iaxcomm, HandleLevelEvent does this:
>    if (input < LEVEL_MIN)
>         input = LEVEL_MIN;
>     else if (input > LEVEL_MAX)
>         input = LEVEL_MAX;
>     inputLevel = (int)input - (LEVEL_MIN);
> 
> to clamp the values and convert them to an integral value for display.
> 
> I'm not sure what observed problem you have with -infinity, or what 
> you mean by runtime singularity error.  But even though I haven't 
> checked this, and I don't remember what the particular rules are for 
> these "special" floating point values, I suspect that the first 
> conditional will always be true as long as LEVEL_MIN is a real number.
> 
> If this was a real problem, we could resolve it by adding 1e-99 to 
> the value before we computed it's logarithm, but I'm not sure it's a 
> real problem yet.
> 
> What is it that you're doing with the value in your callback?
> 

Set the audio meters :-)

I use similar code as in iaxcomm for the clamping of the values.

> >Note that I have added some code to prevent this from happening in
> >vol_to_db(). However, I was wondering whether anyone else had the same
> >problem (i.e. runtime error due to log10(0)) as me.
> >
> >I suspect that I am just not properly setting up the audio output_level 
> >or something along those lines, but I haven't been able to discover 
> >any irregularities with my portaudio audio setup (yet).
> >
> >
> >Eloy Domingos
> >
> >
> >On Mon, 26 Apr 2004 10:52:55 -0400, Steve Kann wrote
> >  
> >
> >>Eloy,
> >>
> >>    Doesn't log10(0) just give you -infinity as a result?  That is 
> >>the correct result for us; the sound level when there's no sound is -
> >>infinity decibels..
> >>
> >>-SteveK
> >>
> >>Eloy Domingos wrote:
> >>
> >>    
> >>
> >>>Hi All,
> >>>
> >>>
> >>>Another thing which keeps bugging me is a problem with the sound system
> >>>of libiaxclient (or better portaudio). I am not sure what the real cause
> >>>for it is.
> >>>
> >>>This is the problem:
> >>>
> >>>In the following function from audio_encode.c, I had to build in a 
> >>>protection to avoid a singularity issue when calling the log10() 
function:
> >>>
> >>>--- snip --- snip ----
> >>>
> >>>static double vol_to_db(double vol)
> >>>{
> >>> //
> >>> // protect the log10 function being called with vol=0
> >>> //
> >>> if (vol<1e-99) vol=0.0001;
> >>> 
> >>> return log10(vol) * 20;
> >>>}
> >>>
> >>>--- snip --- snip ----
> >>>
> >>>The original function just does a straight call to log10(vol) without 
> >>>      
> >>>
> >check.
> >  
> >
> >>>When I traced the call, I got these references of vol_to_db():
> >>>
> >>>lib\audio_encode.c(32): iaxc_do_levels_callback(vol_to_db(input_level), 
> >>>vol_to_db(output_level));
> >>>lib\audio_encode.c(90): volume = vol_to_db(input_level);
> >>>
> >>>In the end it turns out that the output_level=0 an awful lot of times,
> >>>causing the log10() function to blow up.
> >>>
> >>>When I trace the setting of the output_level var, I get these references:
> >>>
> >>>lib\audio_encode.c(7): static double input_level = 0, output_level = 0;
> >>>lib\audio_encode.c(103): calculate_level((short*)audio, len, 
> >>>      
> >>>
> >&output_level);
> >  
> >
> >>>The calculate_level function is called to set both input_level and 
> >>>output_level, and is using the audio stream's data:
> >>>
> >>>--- snip --- snip ---
> >>>
> >>>static void calculate_level(short *audio, int len, double *level) {
> >>>   short now = 0;
> >>>   double nowd;
> >>>   int i;
> >>>
> >>>   for(i=0;i<len;i++)
> >>>     if(abs(audio[i]) > now) now = abs(audio[i]);
> >>>
> >>>   nowd = now/32767;   // note: this seems useless to me
> >>>
> >>>   *level += (((double)now/32767) - *level) / 5;
> >>>}
> >>>
> >>>--- snip --- snip ---
> >>>
> >>>Which would suggest that the audio output can be zero, but the 
output_level
> >>>is still being calculated. For some reason, the input_level does not seem
> >>>to suffer from the same effect.
> >>>
> >>>When I enable the log10() safeguard, the softphone is fully functional,
> >>>without the safeguard, you cannot run the program due to the runtime
> >>>problem of doing a log10(0).
> >>>
> >>>Has anyone seen this problem before? Can it be avoided by certain 
actions 
> >>>      
> >>>
> >or
> >  
> >
> >>>audio settings?
> >>>
> >>>
> >>>Best regards,
> >>>Eloy Domingos
> >>>
> >>>      
> >>>
> >
> >
> >Vocalis Internet Open WebMail (https://webmail.vocalisinternet.com)
> >
> >  
> >


Vocalis Internet Open WebMail (https://webmail.vocalisinternet.com)