Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: looking for high dynamic range samples (Read 91291 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

looking for high dynamic range samples

Hello, I'm working on a software project where I do companding among other things.
I was wondering if there are some decent samples with very high dynamic range.

I found a bunch of test files, where someone speaks at various levels, but I'd like to have some music samples, or something, too.
I don't want to limit myself to Harvard sentences.

The samples should be at least three seconds or so, and don't need to be longer than 10. It doesn't even have to be music at all, it can be just whatever, as long as it's nicely high dynamic range like quiet things and some loud things in the foreground or something.

Re: looking for high dynamic range samples

Reply #1 –
A lot of classical music is highly dynamic, particularly orchestral or piano or organ music.  The 1812 overture is one good-famous example (with the cannon shots), but I'm not really a classical listener so I don't have any other recommendations.  

When I listen to the Les Miserables soundtrack in the car I find myself turning the volume up & down, so I'd say it's highly-dynamic.

In the classic rock genre, maybe Pink Floyd, Dire Straits, or Fleetwood Mac.    But, I haven't measured  the dynamics on any of this stuff, that's just an impression from listening.  

Quote
The samples should be at least three seconds or so, and don't need to be longer than 10.
There are "micro-dynamics", but most loud & quiet passages (in music or movies, etc.)  are longer than a few seconds and such a short selection won't be of much use.


BTW - I prefer the term "dynamic contrast" when speaking about program material.   The EBU R128 standard calls it "Loudness Range".    "Dynamic range" generally refers to the technical limits of the hardware/equipment, or storage/transmission format.   i.e. The CD format has a dynamic range of 96dB (or 93dB?) but the music doesn't have that much dynamic contrast, and you wouldn't want it.

Re: looking for high dynamic range samples

Reply #3 –
The signal-to-noise ratio for 16 bit audio is 96.33 dB (or exactly log10(2^16)*20), but you can get even more perceived SNR out of noise shaping (dithering).

There are plenty of free 24 bit samples and tracks out there on the web that have a large SNR from the quiet on-mic preparation to the full sound, particularly with classical music where you can hear people moving or even breathing.

Schubert String Quartet No. 14, for example:
http://www.lindberg.no/hires/test/2L-093/

I know it's not within 10 seconds, but it does have a lot of range, and you can take the first 6 seconds of it.

Re: looking for high dynamic range samples

Reply #4 –
Quote from: polemon on
It doesn't even have to be music at all, it can be just whatever...
If it doesn't have to be "real world" you can just make some. In a DAW put two tracks with 100dB difference...

Re: looking for high dynamic range samples

Reply #5 –
Exactly what I was thinking.
I don't know of a single hi-rez record that comes with a guarantee of actual high dynamic range. There's always a possibility of high noise floor that makes hi-rez pretty much pointless (probably true for a lot of those recordings). But you can easily generate a very high dynamic range passage digitally by yourself, without introducing any noise above the quantization one.  Audacity will allow you to do that.
 

Re: looking for high dynamic range samples

Reply #6 –
Well the main application isn't music at all, I'm experimenting with companders used in radio transmissions.

Using music would be an easy source I though, since I know there is high dynamic range music out there.

I can upload the samples I'm using right now. One is this: https://www.audiocheck.net/audiotests_dynamiccheck.php
And the other one is just a tone generated in Audacity.

Re: looking for high dynamic range samples

Reply #7 –
Quote from: polemon on
Well the main application isn't music at all, I'm experimenting with companders used in radio transmissions.

Using music would be an easy source I though, since I know there is high dynamic range music out there.

I can upload the samples I'm using right now. One is this: https://www.audiocheck.net/audiotests_dynamiccheck.php
And the other one is just a tone generated in Audacity.

Since there is a LOT of DolbyA material sitting out there -- esp stuff originated in the 1966-1990 -- you can get 10-15dB of dynamic range back before testing your compander by decoding the material.  I can provide decoded snippets of material, or supply the decoder SW that can do *good* things with material on CD (or other sources with the 'taint'.)  I even have stuff from HDtracks at 192/24 that is DolbyA.   Stuff that is DolbyA encoded is often EQed to make it listenable -- the two hard things (not really difficult) are to find the correct EQ to compensate -- usually a simple EQ -- and the correct calibration level.  Both can be found with experience listening, comparing the quality of various samples over an album.  (There is a usual, fairly narrow set of ranges traditionally used for the calibration values on releases, maybe between -12.8 through -15.15dB, more likely -13.8 through -14.80, with highest quality accuracy req of +-0.10dB.)   It seems that with usual calibration values that professional material with tones is in the range of -12.80 through about -13.80.

SO, you can get back the original (mostly low level) dynamics -- with much closer than normal quality using the decoding software rather than a true DolbyA (SW mitigates modulation distortions along with not being dependent on a feedback loop with delays, thereby causing imprecision.)  DolbyA doesn't affect the high level dynamics very much.

We are also working on a Telcom C4 decoder -- it will be INFINITELY simpler than DolbyA for numerous reasons -- even though it seems to be a superior performing design.  (On C4, an encoder/decoder should be of similar complexity and reuse most of the code, unlike DolbyA.)

If you are interested, I can supply a decoder with a 30Sep timeout, or for professional/research purposes contact my project partner at DHNRDS.com for purchasing information.  (I don't do the sales thing -- only talking techie, and trying to help.)

Also, regarding 'fast' companders (those likely to produce modulation distortions), I have some techniques developed over the last few years that can be helpful (and simple libraries of useful intensive SIMD algorithms for dynamics signal processing.)  Some of my techniques use concepts in the somewhat similar to Orban patent US6205225 -- both more CPU intensive, but also more complete (disadvantage/advantage.)  I would need NDA for the anti-MD algorithms, but the SIMD 'helpers' for attack/release type filtering -- gratis.  Mostly that code would just need a bit of docs -- no biggie.

John

Re: looking for high dynamic range samples

Reply #8 –
I found a high dynamics example...  It is called  '05 Dave Grusin,Cripple Creek Break Down.wav' (I decoded it for a friend.)  If the friend allows, I could supply a snippet for demo.  Anyway -- THAT is fairly wide dynamic range, but the hiss might be a little high (I cannot hear it, and the decoding diminishes it significantly.)  Unfortunately, I started with a 44.1kHz, band limited copy.  It isn't so much about hearing, but the original DolbyA could 'see' the >20kHz high frequencies, and so there was possibly a slight/latent amount of HF compression left in the result (I didn't hear it -- but don't know what the recording is really intended to sound like.)

The original on my friends disk was DolbyA encoded with these stats:
             Overall     Left      Right
DC offset   0.000009  0.000009  0.000007
Min level  -0.996338 -0.996338 -0.744141
Max level   0.999969  0.999969  0.720001
Pk lev dB      -0.00     -0.00     -2.57
RMS lev dB    -26.09    -26.46    -25.76
RMS Pk dB     -14.33    -14.33    -14.76
RMS Tr dB     -94.50    -94.50    -94.22
Crest factor       -     21.03     14.44
Flat factor     0.00      0.00      0.00
Pk count        4.50         7         2
Bit-depth      16/16     16/16     16/16
Num samples    12.9M
Length s     293.200
Scale max   1.000000
Window s       0.050


After proper/clean DolbyA decoding, the resulting stats from SOX were as follows:

 
            Overall     Left      Right
DC offset   0.000004  0.000004  0.000003
Min level  -0.822301 -0.822301 -0.579656
Max level   0.951803  0.951803  0.547572
Pk lev dB      -0.43     -0.43     -4.74
RMS lev dB    -29.53    -29.71    -29.34
RMS Pk dB     -16.00    -16.00    -18.20
RMS Tr dB    -108.87   -108.87   -108.33
Crest factor       -     29.13     16.99
Flat factor     0.00      0.00      0.00
Pk count           2         2         2
Bit-depth      24/24     24/24     24/24
Num samples    28.1M
Length s     293.193
Scale max   1.000000
Window s       0.050

Re: looking for high dynamic range samples

Reply #9 –
The compander I'm doing is actually based on graphical algorithms. The idea is to see how a 10-ary deCasteljau can be used to
create a companding function, which can be adjusted "by Ear" when using it over narrow band radio transmissions, like HAM at 10m, etc.

The idea isn't audio excellence, the idea is to see how algorithms that aren't really made for it work in that realm, kinda.
A "low" bandpass isn't a problem at all, and 44.1Hz is way more than what's used commonly over HAM radio. The audio on HAM radio transmissions is almost always band-limited to 8kHz, often less than that.

I'd be quite interested in checking out that high-dynamic range sample you talked about. Doesn't have to be long at all, 10s is more than enough.

Re: looking for high dynamic range samples

Reply #10 –
Okay, the song is "Dave Grusin,Cripple Creek Break Down", the snippet demoed is from 165 to 200secs (long enough for reasonable impact -- maybe a bit longer than you rightfully requested.)  The dynamics seem adequately represented by the high rate mp3, but would use the 24bit/96k flac for anything formal.  The snippets dynamics not quite as high as the entire piece, but still pretty good:

             Overall     Left      Right
DC offset  -0.000114 -0.000114 -0.000016
Min level  -0.844099 -0.844099 -0.634565
Max level   0.910284  0.910284  0.544741
Pk lev dB      -0.82     -0.82     -3.95
RMS lev dB    -26.22    -26.68    -25.81
RMS Pk dB     -12.78    -13.64    -12.78
RMS Tr dB    -131.94   -130.93   -131.94
Crest factor       -     19.64     12.38
Flat factor     0.00      0.00      0.00
Pk count           2         2         2
Bit-depth      29/29     29/29     29/29
Num samples    4.11M
Length s      85.584
Scale max   1.000000
Window s       0.050