This tread has been split off from the DDR thread.
__
Is Low Level Detail an real thing? Are certain systems better or worse at revealing or preserving low level detail in a complex musical, signal? Low level detail goes by other names, such as Micro Detail, Downward Dynamic Range and others. For decades it has been said the some systems are better a presenting this detail than others. This can be considered part of High Fidelity. A Fidelity to the fine details and nuances in the recorded signal.
Mostly this is considered a listener preference, a subjective evaluation of sound quality by trained or experienced ears. It is one of those "You know it when you hear it" experiences. But is there a way to quantify it? Is there a way to test for it, make it audible or show it on a chart? Something that allows us to rank speakers or electronics by their ability to preserve tiny details.
This thread will deal with testing methods that may be able to extract this low level detail from a complex signal and let us hear and measure what is happening in various systems. The test was inspired by Bill Waslo's DiffMaker software and his clever use of a Sousa march hidden 40dB down in a piece of choral music.
In this test we will attempt to extract music and test signals that have been mixed 30dB down into a louder piece of music.The low level tracks are, or should normally be, completely masked by the louder signal. And yet they can be faithfully extracted with the proper software. How well they survive the reproduction chain is the question we will ask, as well as trying to understand what we find.
Test files and a methodology are presented below, as well as test results and thoughts on the relevance of the test and results.
__
Is Low Level Detail an real thing? Are certain systems better or worse at revealing or preserving low level detail in a complex musical, signal? Low level detail goes by other names, such as Micro Detail, Downward Dynamic Range and others. For decades it has been said the some systems are better a presenting this detail than others. This can be considered part of High Fidelity. A Fidelity to the fine details and nuances in the recorded signal.
Mostly this is considered a listener preference, a subjective evaluation of sound quality by trained or experienced ears. It is one of those "You know it when you hear it" experiences. But is there a way to quantify it? Is there a way to test for it, make it audible or show it on a chart? Something that allows us to rank speakers or electronics by their ability to preserve tiny details.
This thread will deal with testing methods that may be able to extract this low level detail from a complex signal and let us hear and measure what is happening in various systems. The test was inspired by Bill Waslo's DiffMaker software and his clever use of a Sousa march hidden 40dB down in a piece of choral music.
In this test we will attempt to extract music and test signals that have been mixed 30dB down into a louder piece of music.The low level tracks are, or should normally be, completely masked by the louder signal. And yet they can be faithfully extracted with the proper software. How well they survive the reproduction chain is the question we will ask, as well as trying to understand what we find.
Test files and a methodology are presented below, as well as test results and thoughts on the relevance of the test and results.
Last edited:
Here are the files
https://drive.google.com/file/d/0B625x1CeTxYuRS1WdXpBWDhuX2c/view?usp=sharing
https://drive.google.com/file/d/0B625x1CeTxYuRHV2M2lsTU5ZUFE/view?usp=sharing
Let me know if you can download and open them
https://drive.google.com/file/d/0B625x1CeTxYuRS1WdXpBWDhuX2c/view?usp=sharing
https://drive.google.com/file/d/0B625x1CeTxYuRHV2M2lsTU5ZUFE/view?usp=sharing
Let me know if you can download and open them
Good, it's all going to be a lot of work, but worth a try. Here is how to proceed. I will try this test further sometime this week.
There is another track hidden 30dB down. Inverting and mixing should reveal the difference, I.E. the hidden track. Thanks to Bill Waslo for the idea.
Best to work in 24 bit or higher, but 16 bit seems OK, too.
- Open the 2 files in an audio editor
- Invert one of them
- Mix the two together
- Listen to the result
There is another track hidden 30dB down. Inverting and mixing should reveal the difference, I.E. the hidden track. Thanks to Bill Waslo for the idea.
Best to work in 24 bit or higher, but 16 bit seems OK, too.
Once you've extracted to hidden file, boost it by 30dB and you will be surprised at how well it holds up.
Next Part: Recording test.
Now you know how well your soundcard will retain buried detail.
Next Part: Recording test.
- Do a loopback connection on your soundcard.
- Playback and record both files
- Do the inversion test to extract the hidden track
Now you know how well your soundcard will retain buried detail.
Real Test Time
Now that you know you can extract the hidden track from the recorded versions, it's time to test drivers!
The idea behind all this is to extract only low level information, the type of info that a bad driver could mask or alter. The better the hidden file comes thru, the more low level or micro detail the driver can reproduce.
Once it is established that you can get a good null from the same file recorded twice, digging out how much of the low level detail gets thru can be done this way. It is important to go thru all the steps at first, to determine how good the null can be thru the recording chain.
Different drivers and speakers can be compared this way, but listening to what they do to information 30dB below the average level.
Now that you know you can extract the hidden track from the recorded versions, it's time to test drivers!
- Record the same file twice from the driver to a mic.
- Trim up and invert (or use DiffMaker) to see if you can get a good null from two recordings of the identical file.
- If not, there is no need to go further.
- If the recording chain is consistent and a null is achieved, we can proceeded.
- Record both files via speaker and microphone.
- Trim up and align the files to extract the hidden track.
The idea behind all this is to extract only low level information, the type of info that a bad driver could mask or alter. The better the hidden file comes thru, the more low level or micro detail the driver can reproduce.
Once it is established that you can get a good null from the same file recorded twice, digging out how much of the low level detail gets thru can be done this way. It is important to go thru all the steps at first, to determine how good the null can be thru the recording chain.
Different drivers and speakers can be compared this way, but listening to what they do to information 30dB below the average level.
Hmmm...
Would it then be possible to quantify the changes made to the low-level signal in more familiar terms (distortion etc), making objective comparisons possible?
University has just started up again, but I'll see if I can find some time to try this.
Chris
Would it then be possible to quantify the changes made to the low-level signal in more familiar terms (distortion etc), making objective comparisons possible?
University has just started up again, but I'll see if I can find some time to try this.
Chris
I'll see what i can do…. i am currently limitedto 16 bit playback and i'd have to find an audio editor.
dave
dave
I don't think the 16 bit playback is a problem, but 24 bit recording and editing will certainly help.
For an editor, Audacity is free and very good. Runs on many platforms. On Windows I've used Goldwave for 15 years or more.
For an editor, Audacity is free and very good. Runs on many platforms. On Windows I've used Goldwave for 15 years or more.
I think so, by embedding a log sweep or something similar 30dB down into the louder signal. Once you have seen that you can extract the low level signal from the recording, you should be able to embed whatever you want as a test signal.
The test works on the electronics chain, I will be trying it ASAP on speakers. A very quiet place is needed.
Going to try testing this on several speakers tonight. Will start a new thread with the results, if there are any.
Tried last night with 4 different powered speakers, from small to large.
Failed miserably. 😡
There was far too much time drift between recordings to get them to line up properly. Even DiffMaker could not realign them. Need to figure out what was happening and try again.
Failed miserably. 😡
There was far too much time drift between recordings to get them to line up properly. Even DiffMaker could not realign them. Need to figure out what was happening and try again.
Are you playing the signal with one sound card / interface and recording with another?
Ideally, you need the play and record clocks to come from a common source. It may help to use the same sound card / interface to simultaneously play back and record.
Ideally, you need the play and record clocks to come from a common source. It may help to use the same sound card / interface to simultaneously play back and record.
Hey Don, yes, same soundcard. M-Audio Fast Track Pro for out and in.
Worked on it this afternoon, got it going good now. It was a software problem. Can now achieve a 96dB null on the soundcard loopback. That's pretty good. Can also pull out the hidden track that is 30dB down with no problem. Sort of amazing. So I think electronics and software are good to go.
Hope to run another set of speaker tests on Monday evening. Keeping my fingers crossed.
Worked on it this afternoon, got it going good now. It was a software problem. Can now achieve a 96dB null on the soundcard loopback. That's pretty good. Can also pull out the hidden track that is 30dB down with no problem. Sort of amazing. So I think electronics and software are good to go.
Hope to run another set of speaker tests on Monday evening. Keeping my fingers crossed.
Just in case you are interested, here are two versions of a file I used for the testing. This is the hidden file that is mixed into the classical piece 30dB down. It is inaudible when listening to the classical piece, but can be extracted and brought back up to level.
One of the files is the original hidden file in a mono mix. The other is the file after it has been:
I have also been able to do this with test tones such as a log sweep. The main change thru my DAC->ADC is the addition of white noise. My guess is that any speaker is going to change it a lot more. How much it changes it compared to others drivers is the important part. We shall see and hear.
One of the files is the original hidden file in a mono mix. The other is the file after it has been:
- Reduced in volume by 30dB
- Mixed into the Copland Rodeo piece
- Output mixed file thru a DAC
- Looped back into an ADC and recorded at 44.1/ 24bit
- Extracted from the mixed recording
- Amplified 30dB
- Trimmed up
I have also been able to do this with test tones such as a log sweep. The main change thru my DAC->ADC is the addition of white noise. My guess is that any speaker is going to change it a lot more. How much it changes it compared to others drivers is the important part. We shall see and hear.
Hmmmm..
Any chance you could extend the test to the amplifier driving the speakers, but not the speakers themselves?
I think a pot across the output of the amp would allow a sensible output level. This would give some idea of how different amplifiers can extract detail, which could be a useful measurement in figuring out why amplifiers can sound different.
Chris
Any chance you could extend the test to the amplifier driving the speakers, but not the speakers themselves?
I think a pot across the output of the amp would allow a sensible output level. This would give some idea of how different amplifiers can extract detail, which could be a useful measurement in figuring out why amplifiers can sound different.
Chris
Yes, I do plan to do that. Perhaps with a simple resistor load, then with a loudspeaker on the terminals. It would be an interesting test.
A very quiet soundcard is needed. Mine is OK for normal use, but the -90dB noise floor rises up to bite me when recording signals that are at -43dB RMS. The noise is obvious. Fortunately everything else seems just fine thru the process. Frequency response and distortion don't change when playing and extracting the signal thru the DAC-ADC.
Want to get the speaker to microphone test working first, then will try some others. Hope to do some speaker tests tonight at work.
A very quiet soundcard is needed. Mine is OK for normal use, but the -90dB noise floor rises up to bite me when recording signals that are at -43dB RMS. The noise is obvious. Fortunately everything else seems just fine thru the process. Frequency response and distortion don't change when playing and extracting the signal thru the DAC-ADC.
Want to get the speaker to microphone test working first, then will try some others. Hope to do some speaker tests tonight at work.
I've worked Pano's files up through a speaker test.
For Buckaroo mix (Rickie) 40dB boost was used to normalize level, correspondingly 40dB was applied to Buckaroo null result.
Attachments are mp3 files with extension changed to .asc for posting.
What do the results show?
For Buckaroo mix (Rickie) 40dB boost was used to normalize level, correspondingly 40dB was applied to Buckaroo null result.
Attachments are mp3 files with extension changed to .asc for posting.
What do the results show?
Barleywater, thanks for doing that! 
What do they show? Well, several things, I think.
What do they show? Well, several things, I think.
- There is a tremendous amount of low frequency noise present in both files. I suppose that comes from air handling or other machine noise. What do you think?
- You got pretty good results aside from the low end noise, the hidden music comes thru well. There is the added white noise that is always going to be a problem with these low level signals, and there is some of Buckeroo still left, which I found in my speaker tests.
- What does that mean? Too early to tell if it is really saying something about the speakers, or if it's more about the test. But it does show that it is possible to extract hidden low end information and listen to it. Comparing speakers with the same set up could be revealing.
- Status
- Not open for further replies.