Randy Yates wrote:> "Clay" <physics@bellsouth.net> writes: > >> On Feb 16, 8:17 am, Randy Yates <y...@ieee.org> wrote: >>> Hi Folks, >>> >>> Does anyone know of any good algorithms, papers, web sites, etc., that >>> discuss ways to determine the perceived loudness of an audio source? You >>> know, like detecting those nasty over-compressed commercials that sound >>> so loud versus the normal program material? >> Hello Randy, >> >> I think the TV commercials may be so overdriven that you may only need >> to detect if the audio waveform has flat tops to it! > > You bring up a good point, Clay. I should probably take at least a cursory > look at some examples and see if it really is that easy.The sharp corners of clipping generate harmonics. Low-pass filtering to restore the mandated bandwidth rounds the corners and hides the telltales. Jerry -- Engineering is the art of making what you want from things you can get. ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Perceived Loudness
Started by ●February 16, 2007
Reply by ●February 16, 20072007-02-16
Reply by ●February 19, 20072007-02-19
Randy Yates wrote:> Hi Folks, > > Does anyone know of any good algorithms, papers, web sites, etc., that > discuss ways to determine the perceived loudness of an audio source? You > know, like detecting those nasty over-compressed commercials that sound > so loud versus the normal program material? > > I'm aware of the basic issues such as peak-to-average power ratio > (i.e., the effect of compression) and dynamic range differences, but > don't know what actually "works" in predicting how "loud" something > will sound to a human.Randy, http://www.replaygain.org/ might be along the lines you are looking for, or at least the loudness measurement part of the standard. It's not an extremely accurate method, but seems to work well in many situations and is not very computationally expensive. Regards Marc
Reply by ●February 19, 20072007-02-19
Randy Yates wrote:> Does anyone know of any good algorithms, papers, web sites, etc., that > discuss ways to determine the perceived loudness of an audio source? You > know, like detecting those nasty over-compressed commercials that sound > so loud versus the normal program material?Frank Klemm developed a way to detect compression in audio signals. You can detect soft and har clipping and compression. I don't know where his website has moved to, but I wrote a program using his algorithm which creates a probability statistic graph from WAV files. It's called 'wavstat' and can be downloaded from my homepage: http://www.huennebeck-online.de/software/download/src/index.html (Source) http://www.huennebeck-online.de/software/download/bin/i686-linux/index.html (Linux Binaries) http://www.huennebeck-online.de/software/download/bin/i386-win32/index.html (Win32 Binaries) The results of 'wavstat' can be seen here: http://www.huennebeck-online.de/musik/stat/index-english.html A very good example how modern remasterings destroy the original dynamic is the Dire Straits' debut album: http://www.huennebeck-online.de/musik/stat/Musik A-E/index.html bye Andreas -- Andreas H�nnebeck | email: acmh@gmx.de ----- privat ---- | www : http://www.huennebeck-online.de Fax/Anrufbeantworter: 0721/151-284301 GPG-Key: http://www.huennebeck-online.de/public_keys/andreas.asc PGP-Key: http://www.huennebeck-online.de/public_keys/pgp_andreas.asc
Reply by ●February 19, 20072007-02-19
Andreas Huennebeck wrote:> Frank Klemm developed a way to detect compression in audio > signals. You can detect soft and har clipping and compression. I > don't know where his website has moved to, but I wrote a program > using his algorithm which creates a probability statistic graph > from WAV files.If you know the old URL, maybe the Wayback Machine has it? Martin -- Quidquid latine scriptum est, altum videtur.
Reply by ●February 19, 20072007-02-19
Andreas Huennebeck wrote:> Randy Yates wrote: > > Does anyone know of any good algorithms, papers, web sites, etc., that > > discuss ways to determine the perceived loudness of an audio source? You > > know, like detecting those nasty over-compressed commercials that sound > > so loud versus the normal program material? > > Frank Klemm developed a way to detect compression in audio signals.Are you sure this was about dynamic range compression in audio? When I Google his name, it always appears in conjuction with audio data rate reduction ...> You can detect > soft and har clipping and compression. I don't know where his website has moved to, > but I wrote a program using his algorithm which creates a probability statistic graph > from WAV files. It's called 'wavstat' and can be downloaded from my homepage:...> The results of 'wavstat' can be seen here:http://www.huennebeck-online.de/musik/stat/index-english.htmlThat's quite interesting - it was another pet peeve of mine to analyse re-mastered and original recordings in a similar way, especially with regard to dynamic range compression. However, I find the cumulative probability view hard to interpret. Wouldn't a normalized histogram (probability density) convey the information in a more direct way. Also, I would subtract the mean and use only a one-sided histogram, with decibels (dBFS) along the abscissa (starting, for example, at -90, up to 0) and density (percentage of samples) along the ordinate.> A very good example how modern remasterings destroy the original dynamic > is the Dire Straits' debut album: > > http://www.huennebeck-online.de/musik/stat/MusikA-E/index.html > > bye > AndreasRegards, Andor
Reply by ●February 19, 20072007-02-19
On Mon, 19 Feb 2007 16:50:03 +0000 (UTC), Martin Eisenberg <martin.eisenberg@udo.edu> wrote:>Andreas Huennebeck wrote: > >> Frank Klemm developed a way to detect compression in audio >> signals. You can detect soft and har clipping and compression. I >> don't know where his website has moved to, but I wrote a program >> using his algorithm which creates a probability statistic graph >> from WAV files. > >If you know the old URL, maybe the Wayback Machine has it?Try http://web.archive.org/web/20040609163457/www.personal.uni-jena.de/~pfk/mpp/clipping.html Regards, Patrick
Reply by ●February 20, 20072007-02-20
Andor wrote:> Andreas Huennebeck wrote: >> >> Frank Klemm developed a way to detect compression in audio signals. > > Are you sure this was about dynamic range compression in audio? When I > Google his name, it always appears in conjuction with audio data rate > reduction ...Yes, I'm sure. I uses his algorithm in my source code. Frank has many projects running ...>> The results of 'wavstat' can be seen >> here:http://www.huennebeck-online.de/musik/stat/index-english.html > > That's quite interesting - it was another pet peeve of mine to analyse > re-mastered and original recordings in a similar way, especially with > regard to dynamic range compression. However, I find the cumulative > probability view hard to interpret.It is quite easy if you know what to look for. Refer to the pink noise diagram at http://www.huennebeck-online.de/musik/stat/Testsignale/pinknoise.png The red line shows pink noise at maximum uncompressed level. Any diagram showing a line which is steeper (to be correct: if the values are more negative on the negative side and more positive on the positive side) than the red line (e.g. the light blue line) indicates compression. Hard clipping can be seen at "http://www.huennebeck-online.de/musik/stat/Musik A-E/Marie Boiune - Leahkastin.png" where the maximum level is never reached. An algorithm checking for consecutives data values at maximum level would fail utterly. Soft clipping can be seen at "http://www.huennebeck-online.de/musik/stat/Musik A-E/lJ.S.Bach - The Goldberg Variations, BWV 988, Jean Guillou, Kleuker Organ Alpe d'Huez1.jpg" The dark green line reaches a value around -31000 at sigma=-4 but does not stay there for more negative sigma values.> Wouldn't a normalized histogram > (probability density) convey the information in a more direct way.No, this does not work. I tried ...> Also, I would subtract the mean and use only a one-sided histogram,In principle yes, but there are compression algorithms which create nonsymmetrical output and this could no longer be detected.> with decibels (dBFS) along the abscissa (starting, for example, at > -90, up to 0) and density (percentage of samples) along the ordinate.Feel free to do so, the source is GPL. bye Andreas -- Andreas H�nnebeck | email: acmh@gmx.de ----- privat ---- | www : http://www.huennebeck-online.de Fax/Anrufbeantworter: 0721/151-284301 GPG-Key: http://www.huennebeck-online.de/public_keys/andreas.asc PGP-Key: http://www.huennebeck-online.de/public_keys/pgp_andreas.asc
Reply by ●February 21, 20072007-02-21
Patrick Piecha wrote:> http://web.archive.org/web/20040609163457/www.personal.uni-jena.d > e/~pfk/mpp/clipping.htmlThanks! Martin -- Quidquid latine scriptum est, altum videtur.
Reply by ●February 21, 20072007-02-21
Martin Eisenberg <martin.eisenberg@udo.edu> writes:> Patrick Piecha wrote: > >> http://web.archive.org/web/20040609163457/www.personal.uni-jena.d >> e/~pfk/mpp/clipping.html > > Thanks!Thanks to all who responded. I'm swamped at the moment but hope to get back to this in the next couple of days. -- % Randy Yates % "She has an IQ of 1001, she has a jumpsuit %% Fuquay-Varina, NC % on, and she's also a telephone." %%% 919-577-9882 % %%%% <yates@ieee.org> % 'Yours Truly, 2095', *Time*, ELO http://home.earthlink.net/~yatescr
