Digital Sound Processing - DSP (up-encoding analysis), page 4

SpasVstar V.I.P. on March 7th, 2009 / post 28995
:-)
For curiosity only:
I have tried LAME (3.98.2) -q0 -ms -V0 and two different files: Transitions vol. 4 2008 (72 min) and my enhanced XM rip of John Digweed's part of Transition 01-Mar-2009 show (54 min).
Obviously the source files are different but nevertheless...

LAME applies (at least) a 19 kHz low pass filter in VBR mode of compression.
The result I have got:
Transitions vol. 4 2008 - 267.6 kb/s
Transition 01-Mar-2009 - 262.3 kb/s (262.3/267.6 = 0.98)

Then I have tried LAME (3.98.2) -q0 -ms -V0 -s32 with two different low pass filters and my enhanced XM rip of John Digweed's part of Transition 01-Mar-2009 show:

The results I have got:

--lowoass 16: 234.7 kb/s or 3.667 bits/sample vs 2.179 bits/sample with 192 kb/s 44.1 kHz
--lowpass 15: 216.2 kb/s or 3.378 bits/sample vs 2.000 bits/sample with 192 kb/s 48.0 kHz

For the additional 1 kHz bandwidth from 15 KHz to 16 kHz LAME has increased the bit rate with 8.56%.
For additional 22.2% file size increase (16 kHz band width) there is 68.3% increase of information to recover the sound. It is the best solution, at least for me, when considering the sound quality of an enhanced SBD FM radio source.
Skype:spas.velev
madesstar house addict on March 8th, 2009 / post 29004
1) You didn't give any reply to the fact i showed, that Sirius Rips are heavily sound enhanced and therefore they look so nice on the graphs but the listening quality drops. In fact, you do not comment ANY of my valid graphs, instead you keep posting your graphs with just one argument and that is dB drop above 16Khz. But as i understand correctly your graph, it shows the AVERAGE volume across the spectrum bandwidth.  And if even 1second of the file is above 16Khz, it cannot be FM or 128k.
2) The producer doesn't distribute lossless copy, none i know of who has large syndicated show (AvB, Markus Schulz, ..). There is no reason to, it would be more complicated regarding bandwidth, etc, not to mention that many tracks a dj plays come from a 320k mp3 source.
3) It is easy to enhance a sound to look like on your graphs. But it will be in fact unlistenable. (this one im not 100% sure about, but i managed to edit the file to look like yours in Adobe Audition by adding the volume and noise to higher frequencies. But i might be wrong.)
4) Original source cannot be Kiss Fm broadcast because they broadcasted it later than it was available here! (because you keep repeating it only based on your graphs and not objective facts)
5) SBD means SOUNDBOARD.  According to mp3 rules:
"MP3 FILE supplied by a radio station or DJ (and not recorded from a webstream)"

p.s.: Why 320/192k look the same. Well, they obviously come from the same source: The source DJ himself supplied. They use different codecs with different settings, yet they look exactly the same !
p.s.2: I will repeat again: They cannot be from Kiss Fm because they were both available sooner than kiss Fm broadcast happened.
SpasVstar V.I.P. on March 9th, 2009 / post 29023
:-)
Thank you for showing me the meaning of SBD in the context of the Official MP3 Release Rules 1.1.
With my respect to the rules I have already found at least one exception related to the SBD.

First, the fact we agree on.
“p.s.: Why 320/192k look the same. Well, they obviously come from the same source:…”  and your interpretation follows.
Here is my interpretation. For simplicity I’ll assume not only “320/192k (spectra) look the same” but they are equal also which is true (P.S. for those moments of time we inspect the spectra. To be true for the whole file we need to inspect the spectra at all most all every moment which of course is impossible - that is why the listening tests are conducted.) within the limits of the resolution we have.
The spectra have strict mathematical definition and are calculated based on digital (PCM) sound data. If two spectra are equal their sound data are equal also.
The data for the calculation of a spectrum of a sound which digital data are mp3 encoded are obtained by decoding the mp3 data. This operation is strictly defined by mp3 standard (MPEG-1 layer 3). If the sound data are equal so equal are the mp3 data also.
Or in other words, the TALiON 192k mp3 file contains the same digital information as the Proton 320k mp3 file. The only difference is the Proton 320k mp3 file contains non-significant zeros. (for example - as 09 instead of 9 – two digits instead of one).

There is one question we were not agreed: what the source of the show is?  Or you have already changed your mind. You said Proton 320k was the source and hence the conclusion - there exists not a better sound for the show because 320 kb/s mp3 encoding ensures the best possible mp3 compression. This implies no other encoders are used which obviously, for me, is not true. Besides the lossless coding the current international audio coding standard is MPEG-4.
Besides, TALiON 192k and Proton 320k are 16 kHz bandwidth 192 kb/s mp3 encoded sounds. If they were not 192 kb/s then both spectra should be different. A 16 kHz 192 kb/s mp3 sound obviously isn’t the best possible mp3 encoded sound.

Further, there are assumptions in your post I don’t want to argue about.
Finally, two more things to mention:
1) “… you do not comment ANY of my valid graphs, instead you keep posting your graphs…” If you don’t see the difference between graphs you obviously don’t understand me. I don’t know if it makes any sense for me to try to explain. But nevertheless … what I tried to show and compare was the sound power distribution at some moment of time, within the time resolution and to evaluate its closeness between the spectra. The closeness is difficult to evaluate but it was possible to see “the same” power distribution and lacking of power at all. That is impossible to do using your “valid graphs”.
“AVERAGE volume across the spectrum bandwidth” has no meaning for me at all and I don’t want to go in a topic such as a human auditory perception which I am not familiar with. What is clear though is the perceptiveness is frequency and time dependent and such characteristics like AVERAGE over frequency and time do not have meaning, at least for me.
2) “Sirius Rips are heavily sound enhanced” - it makes no sense for me to discus the XM radio communication channels. Such a discussion would not be constructive and would be based on not well reasoned assumption and guesses. As to your discovering the difference at around 21:55 of the show, yes there is a difference. It is a regular sound, not a random noise. Why it is missing in the 192k version I don’t know. Don’t ask me about that. What I know is this sound is present in the XM broad cast and it is better heard in my file due to my processing.
Skype:spas.velev
madesstar house addict on March 9th, 2009 / post 29024
Just one thing:
"It is a regular sound, not a random noise"

I did show you it is NOT THE REGULAR SOUND BECAUSE THE ORIGINAL TRACK DOES NOT HAVE IT ! That pretty much disqualifies all of the Sirius rips in my eyes. If you want to discuss technical aspects of a sound like spectrum bandwidth you HAVE TO consider also broadcasting quality and their techniques for broadcast.

Your graphs show VOLUME distribution across the spectrum bandwidth. (one axis dB, another one Hz)
My graphs show SPECTRUM distribution across time. (one axis Hz, another one time)

Your graphs are misleading when considering the encoding bitrate, because they render the sound "valid" (eg not approximation of 128k) only if there are enough strong volumed high frequencies existing. So if there would be "beat only" track with no frequencies lets say over 10Khz, you would say its 96k source even though it could be lossless format. The 96k might be sufficient for encoding but NOT the source.

as for other things (like replying, its "16Khz Bandwidth" even though it goes over 16Khz on regular basis, even though the original Digger CD looks very similar when encoded, even though your rip is so heavily enhanced it shows 20Khz sounds on a part where there are ONLY BEATS, etc), i think it has no sense to repeat what was already said and shown.
TomMixlightning 3daywarning on September 29th, 2009 / post 31089
:-D here we go again!

big thanx to mades for the sample with the 'sirius enhanced' sound!

that is what i am talking about for 3years now:
sirius is sending a proven bandwidth of less than 64kbits per second to spasv's crappy sirius-decoder
he does a whatever-so-secret-decoding.
it is a proven fact that the sirius data is artificially altered/added/blownup to enhance the sound
for the paying listener.

so spasv is recording it and processing it...
making a 300kbit/s blownup bandwidth-chopper  :o)
spasv has the midas touch  :whistle:
SpasVstar V.I.P. on September 11th, 2011 / post 41529
Hi,
I thought to make a post about the sound quality of some sets I have seen here but before that I saw the ignorant comments about my work and I would like to say this for the last time.

Sirius radio broadcasts a sound with a spectrum width of 15 kHz.
You can easily check this with any TALiON's rip from Electric Zoo Festival New York 09/2011, for example.
Just listen to hear and make sure they were broadcasting through Sirius and then run a Spectrum Analysis of the sound file.
You'll see a spectrum cut at 15 kHz.

The MediaInfo would say about the encoder:
Writing library                  : LAME3.98r
Encoding settings                : -m j -V 0 -q 0 -lowpass 19.5 --vbr-new -b 32.

Then encode ANY CD sound with LAME 3.98 and parameters: -mj -q0 -b95 and check its spectrum out.
(-b95 tells encoder to use Constant Bit Rate at 95 kbps.)
You'll see the same sound bandwidth of 15 kHz.
(Here is what I got:
>lame -q0 -mj -b95 airscape.wav
LAME 3.98.4 64bits (https://www.mp3dev.org/)
CPU features: , SSE (ASM used), SSE2
Resampling:  input 44.1 kHz  output 32 kHz
Using polyphase lowpass filter, transition band: 15097 Hz - 15484 Hz
Encoding airscape.wav to airscape.wav.mp3)

So, one could conclude Sirius is broadcasting sound filtered at 15 kHz which correspond to a sound quality as if it was mp3 encoded at 96 kbps.

BUT as to me,
I have been encoding the sound of another satellite channel which former name was The MOVE of XM satellite radio. Sirius was a satellite radio also. Both radios have merged in 2008 and the new satellite radio name is Sirius XM. The former MOVE channel broadcast under the name AREA and is completely different from Sirius division. At least it was using different method of encoding the sound.

Beside that, I have reconstructed a CD spectrum of the Area broadcasts using my own filters - programs I have written implementing FFT, if this means anything to you.

Finally, I have seen many 320 kbps mp3 rips, even here also, with a spectrum of 15-16 kHz.
The LAME encoder filters 320 kbps at 20 kHz and -q0 -V0 (the best variable bit-rate encoding) at 19 kHz. This means those rips have been re-encoded from a band-limited mp3 at 95-128 kbps or FM broadcast source.
If you re-encode a lossy encode you'll get an APPROXIMATION to your source and you can get the source quality if you only make a lossless encode.
So, it doesn't make any sense to encode an 95 kbps or 128 kbps at 320 kbps only to get the closest mp3 encode to your source.
IT IS MUCH BETTER SIMPLY TO RECORD THE SOURCE mp3 STREAM.
Skype:spas.velev
SpasVstar V.I.P. on September 24th, 2011 / post 41652
More about the sound spectra that could be helpful. :-)

First of all, what is a sound spectrum?
The sound is a wave of pressure we hear when it propagate through the air. A sensor-microphone can generate an electrical signal to mach exactly the pressure wave. This electrical signal I an analog signal meaning it is determined at every moment of a time range and its value is every value in a range of values.
The computers dont process such signals. Instead, they work with discrete time signals. A discrete time signal is determined at discrete moments  of time in some time range, so these moments are finite number and its values are taken from a finite set of values, for example the set of all 16-bit binary numbers. Such signals are represented by discrete time functions. We will call the discrete time functions that represent some sound wave digital sound. In order to be played a digital sound needs to be converted back to analog signal.
There exist strict mathematical transforms which can transform a function of a time, like a function representing an analog sound signal, to a function of parameter, called frequency. Based on such transform are functions called spectra. Among them is the Power Spectrum or simply spectrum which Im going to use.
The scientific researches have establish that a human can percept sounds whose spectrum parameter is in the range 20 Hz 22 kHz.
According to this the Red Book the Sony-Philips standard for digital recording of audio CD a digital sound needs to be generated by sampling the analog sound signal at a rate of 44,100 samples per second and the sample digital value needs to be a 16-bit digital number. 44.1 kHz sampling rate has been chosen based on the theory stating that an analog signal can be perfectly reconstructed from an digital time signal if the analog signal is band limited and the sampling rate is at least twice as high as the band limit.
There are discrete time transforms to process the discrete time functions and we are using them.
Finally, as long as the transforms are strict mathematical operations everything one can conclude from a transform (spectrum) applies to the discrete time function (digital sound) also. In particular, if one conclude that two spectra are close then the digital sound functions are close also.

All mp3 spectra shown in this post are obtained using LAME 3.92.4 encoder.

Now, there is a spectrum shown on the figure below. It has two views: in a logarithmic scale, which allows for the low frequency range to be observed and in linear scale, which actually hides the low frequency range.
The spectrum is calculated based on the Tiestos remix of the Delerium [Featuring Sarah McLachlan] Silence.
It is a CD rip.  So, you can see the spectrum is band limited with upper limit of 22.05 kHz.
This spectrum is of a quality sound:
It has almost flat area until frequency 10.5 kHz, then slowly declines until 20 kHz, and there is a transient area 20 22 kHz.
A worse quality sound usually is below this one. Of course, it depends on the sound itself.
Important things to remember:
- all audible (20 Hz 22 kHz) frequencies are present in the spectrum.  The spectrum bandwidth is 22 kHz.
- There is a simple sound quality rule saying the sound quality is the bandwidth.

 
Fig 1

Further on Im going to use only the linear scale representation because the changes Im going to show are easy visible in the high frequency  spectrum range.
Next figure shows five spectra:
Lossless it is the original CD rip,
Next four are from mp3 (-q0) default encoded at 320 kbps, VBR max quality ( V0), 192 kbps, and 128 kbps digital sounds.


Fig 2

As you can see its almost impossible to draw some conclusions based on the differences seen the low and middle frequency range. And Im not interesting in them. So, further Im going to show the frequency range Im going to focus.


Fig 3

The simple bandwidth quality rule here:
320 kbps 20 kHz, 251 kbps (VBR) 19 kHz, 192 kbps 18.5 kHz, 128 kbps 16.5 kHz.

Next, Im going to discus two sounds:
01-gramatik_-_live_at_the_electric_zoo_(new_york)-sat-09-04-2011-talion.mp3 you can download it from tibalemixes.com
01_gramatik___live_at_the_electric_zoo__new_york__sat_09_04_2011_talion.mp3 you can download through a link provided by mixing.dj
Because they both have the same name Ill rifer them talion and mixing.

First, lets see the talion release.
If you take information about the file using MediInfo  youll get:
Encoding settings: -m j -V 0 -q 0 -lowpass 19.5 --vbr-new -b 32
This is exactly the default settings  lame is using if you start it this way: lame q0 V0. Lame adds the rest.
Here is its spectrum along with three more as references.


Fig 4

The talions spectrum is below the others obtained from the CD rip. It is different sound indeed.
As it is seen talions spectrum is band limited at 15 kHz while lame low pass limit imposed by lames filter would be 19.5 kHz. It's band limit is 15 kHz and this is because their source is band limited.
As I have already said in my previous post, their source is Sirius channel of the satellite radio Sirius XM and it is some kind of FM radio. (Analog FM radios are band limited to 16 kHz.)
Because TALiON dont specify the low pass filter to be used by LAME it uses it's default with a band limit of 19.5 kHz. Because of that it uses 44.1 kHz as a sampling rate, also. And a sample is encoded in average with 216 kbps/44.1 kHz = 4.978 bit per sample. Had they used parameter lowpass 15 or resample 32, LAME would use 32 kHz as a sampling rate. With such a sampling rate lame vbr encoded Silence (Tiesto remix) has 201 kbps which gives 201/32 = 6.28125 bit per sample in average.  Having in main that the source CD rip uses 16 bit/sample it is obvious a 32 kHz sampled encode would be of better quality. (The lower bit per sample introduces so called quantization noice.)
Besides, this is true for any analog FM radio. They broacast band limited at 16 kHz sound. So, the perfect encode would be sampled at 32.0 kHz.

Finally, lets see the differences between talion and muxing.dj releases.


Fig 5

I would say there are not visible differences. You can see them in the transient range but there they are not important at all.
But nevertheless there are differences. The quality of a re-encode, if no special processing has been made, is worse than that of the its losy  encoded source. The best re-encode possible would be a lossless encode which will perfectly reconstruct its losy source. But this is meaningless.

So, why is that? The reason is as follow.
The re-encode is made using a decoded losy encoded source. The chances that the encoder will throw the same information as during the encoding of the original source, even it uses higher bit-rate, is ZERO. Thats way the encoder will throw different information which means it will increase the information lost from the source.
So, I don't see any good reason for the TALiON's release to be re-encoded @320 kbps. The result is increased file size and lower sound quality.
Skype:spas.velev
slash ProDanceCulture on September 24th, 2011 / post 41653
thanks alot, SPAS, for doing this all! with charts and diagrams people must trust this better, than when i said "mixingdj is upencoding their mixes". noone cared. now i hope some people will open their eyes!

for those, who doesn't see the idea of this article... it is here to show, that files some people download from mixing.dj site are A WASTE OF YOUR BANDWIDTH! mixing.dj is out there to have you pay for your downloads, but they don't do it directly, they want you to download huge upencoded files so that you'd spend your daily/hourly whatever limits with the file sharing host where they store the upencoded mp3s, and in the end you can't take all the waiting and pay $10 for the fast downloading access with that file sharing host...

i hope you see the difference now... sites like tribalmixes, themixingbowl and maybe 2-3 others provide FREE MUSIC for FREE with no (or very limited) advertising. while sites like mixing.dj and similar blogs do it for money, they get percentage and stuff, they put so much advertising on their site, that you can hardly click the right link without triggering some add... this is truly sad... and i really hope everyone will realize at some point, that FREE stuff must be FREE.,,


good luck!
you cannot post in this forum.
click here to to create a user account to participate in our forum.