I think may be the small detail is being artificially enhanced by highly distorting SETs.
An externally hosted image should be here but it was not working when we last tested it.
(I hope this works and the image is from and presumably owned by Rod Elliott of ESP.)
The way waveforms add it is the higher frequencies which clip first.
In the case of SETs if I remember correctly that is mostly 2nd harmonic and generally even order which is musically related to the 1st harmonic (aka fundamental). Basically a new signal not present on the recording is generated one octave up from the original and the 4th harmonic would be two octaves.
Musically the equivalent would probably be playing chords when the composer asked for single notes.
As for newly discovered details and just speaking for myself once I hear a previously missed detail when replaying a tune on another system I hear it all the time on most systems. Probably not as clear or obvious but it is still there usually.
Hi Mitch, I just ran into this post. Something in the DiffMaker help files tends to be almost always overlooked. The length of the tracks sampled should usually be on the order of WELL under 60 seconds (I usually used 10 seconds), because the memory required and the likelihood of complications gets astronomical quickly. DM oversamples (a lot), does scads of transforms, comparisons, and all that, so memory gets eaten up at a rate on the order of many video editors. Besides, for comparisons done without locked record/play clocks (on all signals), the biggest difficulty is usually sample rate wandering. DM handles to about 3rd order wobbles (such as a "increasing, decreasing, then increasing again" case), but beyond that all bets are off. 60 seconds of recording with unlocked clocks is unlikely to vary in such a simple way and it only takes a ridiculously small sample time error to blow the whole deal. Also the frequency response correction trick (EQ) it can do doesn't do well with rate variations of any type, really.
(BTW, I'm not developing DM anymore since about 7 years, it did what I needed it to do then for my own investigations, and further work or support isn't really in the cards. But thanks to all for trying it!).
(BTW, I'm not developing DM anymore since about 7 years, it did what I needed it to do then for my own investigations, and further work or support isn't really in the cards. But thanks to all for trying it!).
Hey Bill, thanks for your DM software! Many folks have used it with good success. Sure, understood on the 60 seconds. I have been getting reliable and repeatable results with around a 30 second test file as did Archimago. It has been fun! And very revealing
Have you considered open sourcing the software on GitHub? As a fellow software developer, I know other software developers and audio enthusiasts would be interested in carrying the DM flag forward, if you fancy that.
All the best, Mitch
I let this drop because of a sudden move of location of about 8000 kilometers.
I did something with my mixing desk a few years back. It has multi-track recording capabilities, so it's easy to loop and re-loop things - just move the slider back to the start and you've got the same section of pink noise over and over again. Handy.
So how about this as a test:
- Record a section of pink noise within the desk (that way, we're using the same bit each time)
- Play that pink noise through a speaker while recording the mic input
- Seperate channel - mix the pink noise with a swept sine, or whatever else
- Play noise + sweep through the speaker, record the mic input
- Import into Audacity
- Invert one recording, and sum
- Ought to leave you with a sine sweep
- Import into REW
- ????
I think the think to do there would be to also record the speaker playing the swept sine tone without pink noise, and then you've got a comparison - low level sweep vs low level sweep plus high-level noise.
From what I can tell, that's holding as many things constant as possible - all we're doing is introducing (repeatable) pink noise while doing a sweep.
I'd probably do it close-mic'd so SNR improves with regards to outside acoustics, noise, etc. Anything that gets recorded will screw up the measurements.
Not really in a position to do much audio-related stuff right now, but once things settle down this is something I'd like to get back to.
Chris
That seems like a good idea Chris. I like the idea of recording the low level signal alone as a comparison. 
The main stumbles I found were getting the files lined up at the correct staring point and also getting them level matched. Level isn't too hard, if you record one after the other.
Yes, close miking is required if you don't have an anechoic chamber. We are really just concerned with how well the driver can reproduce low level detail. In this test it is low level detail buried inside a higher level signal - but maybe just a very low level comparison of a single signal would tell us a lot.
The main stumbles I found were getting the files lined up at the correct staring point and also getting them level matched. Level isn't too hard, if you record one after the other.Yes, close miking is required if you don't have an anechoic chamber. We are really just concerned with how well the driver can reproduce low level detail. In this test it is low level detail buried inside a higher level signal - but maybe just a very low level comparison of a single signal would tell us a lot.
A very interesting topic.
I have noticed a fading detail effect. When a single driver gets increased lower frequency power (and movement) it appears to diminish the detail of the higher + lower amplitude frequencies. It looks like superposition problem. I was never sure if it was me or the driver.
Would the level of detail the driver is capable of be indicated by it's HD (THD or even harmonics ) ?
I have noticed a fading detail effect. When a single driver gets increased lower frequency power (and movement) it appears to diminish the detail of the higher + lower amplitude frequencies. It looks like superposition problem. I was never sure if it was me or the driver.
Would the level of detail the driver is capable of be indicated by it's HD (THD or even harmonics ) ?
Hey Pano,
A lot of that can be sorted within the desk - move the recording slider back to the start, press "track" on the pink noise and "record" on the mic input, and I'll be able to subject the speaker to the same section(s) of pink noise as many times as I like, with other things mixed in as needed.
In theory, that'd get everything to work perfectly.
In practice, I expect the increased noise levels present in the difference file to swamp any changes in the distortion from driving the speaker with other signals.
We'll see.
I suppose this'd be a good way to test amplifiers, too.
Chris
Sounds like a fun game.
For power amps, we could take it up a notch - low-level detail vs load impedance vs voltage output.
... and now I need to build another load bank.
Chris
Extracting a signal from under a lot of pink noise is the test so far.
I was thinking the test signal could be swept sine, which can then be imported into REW for analysis.
Chris
I used Reaper with ASIO in multitrack mode, track one playing the signal and subsequent individual tracks recording the loopback events.
Comparisons of recorded tracks showed automatic perfect time alignment.
Dan.
Picking this up again, and still not getting the results I was hoping for.
I was wondering just how much detail is "thrown away" by the MP3 conversions at different sample rates. Inspired by Bill Waslo's brass band that is hidden 70dB down in another file but can be extracted with Diffmaker software, I set out to do a similar test of MP3s.
Basically I took a Frank Sinatra track and buried a classical wind ensemble 60dB down in the mix. When this is saved as a 24 bit file, the extraction of the very low level wind ensemble is rather good. When saved as a 16 bit file, it's not so great and rather noisy, but still there, recognizable and surprising, considering how buried the signal is. The hidden signal lies at an RMS value of 85dB below peak. That was my proof of concept and it works.
From there I moved on to MP3 encoding of the Sinatra song with hidden wind quintet. But even at 320 Kbps what's left over after extraction is so loud it hides the wind quintet. In other words, the difference between the wav file and the MP3 is so great that it still hides the low level signal.
That leaves me at an impasse. Is there perhaps a way to further extract the hidden signal and pull it out of the noise?
I was wondering just how much detail is "thrown away" by the MP3 conversions at different sample rates. Inspired by Bill Waslo's brass band that is hidden 70dB down in another file but can be extracted with Diffmaker software, I set out to do a similar test of MP3s.
Basically I took a Frank Sinatra track and buried a classical wind ensemble 60dB down in the mix. When this is saved as a 24 bit file, the extraction of the very low level wind ensemble is rather good. When saved as a 16 bit file, it's not so great and rather noisy, but still there, recognizable and surprising, considering how buried the signal is. The hidden signal lies at an RMS value of 85dB below peak. That was my proof of concept and it works.
From there I moved on to MP3 encoding of the Sinatra song with hidden wind quintet. But even at 320 Kbps what's left over after extraction is so loud it hides the wind quintet. In other words, the difference between the wav file and the MP3 is so great that it still hides the low level signal.
That leaves me at an impasse. Is there perhaps a way to further extract the hidden signal and pull it out of the noise?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.