Le Wed, 16 Feb 2011 19:51:09 +0200, Alfs Kurmis <kallipy...@inbox.lv> a écrit :
> > Hi Experts. > I wanna normalize my sound stream by loudness (energy / pressure / > intensity) , not by peaks. > How i do it ? > Is available Jack plugin for so what ? > What is (we hear as) "loudness" ? Loudness is an adaptive filtering that depend on the position of the volume knob. At full volume, no filtering is done, at low volume, the low and high frequencies are more amplified than the medium ones. > RMS or +(average) or something else ? In Audio, that is a mean to express or measure the sound volume. http://en.wikipedia.org/wiki/Root_mean_square It is different ways to measure the sound level. For low end home equipment, the sinus power is used. Such a sound doesn't correspond well to a real life situation, that is when your are listening to some music. It can also be very stressful for an amplifier. A much better way to express the power of a sound equipment, is to use some kind of white or pink noise instead of a sinus. That is how the rms power is measured. In such, it doesn't tell anything because you have to know the condition of the measure (usually the characteristics of the white or pink noise, and a given rate of distortion) in order to be able to interpret it. In practice, the rms power is usually a little bit lower than the sinus power measured at the same rate of distortion. > Is somewhere available examples how to calculate RMS ? > Is it done simlpe by : >  int i, n;  double sums, rms; >  sums=0.0;  n=10;  rms=0; >  for(i=0; i<n;i++) >  {   sums = sums + ( (double)i * (double)i );  } >  rms = sqrt(sums / n); >  printf("          > rms = %12.12fnn", rms ); > Is so sipmple algo good enough for frequencies > 10 kHz ? > How to calculate RMS with hi-precision for frequencies > 10 kHz ? > With inerpolation or so ? > What is reference ( 0dB ) RMS for example for 16bit PCM signal 1024 > samples ? >   sqrt( 1024*(0x7FFF ^ 2) / 1024 )  ==  sqrt( (0x7FFF > ^ 2)  ) == 0x7FFF  > Is it simple  0x7FFF (32000 dec) ? > How to calc RMS for stereo signal ? > So , or somehow else ? > for(i=0; i<n;i++) >  {  .... >    sums = sampleL/2 + sampleR/2; >  } > In case operating with float point, what should be a bit > faster,  / 2  or * 0.5 ? > How to gotta RMS value in dB ? >    20 x log10(Measured/Reference0_dB)  > or 10 x log10(Measured/Reference0_dB)  ?? > Im just physics student and im new in DSP, > so pleaz no angree about simple or stuppid questions. > Any examples and hints welcomed. > Many Tnx in advance. > Alf > ---- RMS is used in audio to express the power. By analogy, some manufacturers are using it to express the output or input voltage of their equipments. But it doesn't change anything, the math is always the same with the db, When you are talking about power, you have 10 * log (P1/P2). When talking about voltage or current, it become 20 * log (U1/U2) = 20 * log (I1/I2) http://en.wikipedia.org/wiki/Decibel The 0 db of an audio equipment is the maximum level it can handle at a given rate of distortion, and with a given and constant waveform or noise. With the analog electronic, your equipment will be able to handle higher values of the input signal during a short period of time (in some cases even much higher values). This is what we call the headroom. It can be expressed by the measure of the musical power, it is the same issue than with the measure of the rms or sinus power, you have to know the exact condition of the measure in order to be able to interpret it. In other words, the specifications must tell you enough in order to be able to reproduce exactly the same measurements and all the described measurements. If it is not the case, they are just bullshit. Another issue is that it is no normalised way to measure and express the specifications of an audio equipment. Each manufacturer is using its own measure protocol. So, it can be difficult to compare them on the paper. And many specification sheets are just bullshit because they doesn't tell enough about the conditions of the measurements. With a digital signal, you will get no headroom at all. The maximum is when all the bits are to 1. If you add something, the bit number doesn't change and you will get a wrong result. It is one point on which jack with its floating point approach is better. You can do calculations inside jack that will be outside the scope of the sound card but valid inside jack. With a floating point approach, you can define the 0 db to be somewhere on the middle, and do calculations with a good approximation, calculations that would be impossible in practice with an integer based system (because of the added complexity to take the errors in account). Of course, you will still have to insure that the d/a converters of the sound card will handle the result of your calculation. Such converters know only about integers, if the digital sound level inside jack is too high, it will be clipped by the sound card. Inside jack, all you will have to do is to be sure that the result of your calculation will be inside the limits of jack. When outputing to the sound card, a mixer and a vu-meter are good things to have. Dominique -- "We have the heroes we deserve." _______________________________________________ Linux-audio-dev mailing list Linux-audio-dev@lists.linuxaudio.org http://lists.linuxaudio.org/listinfo/linux-audio-dev
