I probably should read the papers, but if measured data can be explained in physical terms, fidelity can be well understood. Then it comes to which physical terms of measured data are more critical is probably key to an improvement development cycle, which is probably where the metrics will come to play.
Dear mbrennwa,
Just to clear up some misconceptions per your recent reply to me, I provide the below thoughts:
- You made a statement that to allude to the fact that it is “my software,” which is not true in any sense of the word. As I have made clear via my 1st posting to this MI Pro GB forum and as Virtins posted in this MI Pro GB forum, I am not an employee, consultant nor in any way affiliated with Virtins in any official capacity (same with RTX Corp.)
- My main reason for getting involved with this whole MI Pro GB is that I had seen the need to have the RTX6001 unit have a custom API in order to have true autoranging and autoscaling, (just like Audio Precision gear does). To achieve this goal required I put on the hat of “coordinator / facilitator” and proposed ideas to both Virtins Technology as well as RTX Corporation. I arranged for the transfer of MI Pro USB dongle licenses to be transferred from Singapore (Virtins) to Denmark (RTX) and conversely for one RTX6001 with all mods installed and the latest firmware update applied to be shipped from Denmark to Singapore. Once this was accomplished we had to see if the “theory” of making a custom API could manifest itself into reality (an actual API capability between MI Pro and the RTX6001). I’m happy to say that prove to be the case in point.
- Once the above was done, I agreed to run a diyaudio MI Pro GB for Virtins needed a return on investment (ROI) for their time to code an API, etc. That ROI was the guaranteed sale of at least 42 MI Pro licenses. To sweeten the deal, Virtins has allowed me to offer a discounted pricing scheme on MI Pro licenses based on the total number of MI Pro licenses sold (per my posting # 1 to this GB) as well as to be able to offer a 10% discount on Virtins add-ons or the Virtins “full package” the latter which has all the MI add-on modules included.
- Before all the above happened I had pushed Virtins to include other features I wanted, (remember, I have personally purchased the MI “full package” version many years ago from them). So yes, I coordinated things between Virtins and Dr. Earl Geddes to ensure that the GedLee Metric (Gm distortion) and some other features I wanted were included into Virtins. I can only tip my hat to Virtins and give a big “thank you,” for working with me to implement things into MI I desired.
- Let’s also define the terms a bit. You continue to use the term “NCM” when you met to write “NCD” (it is called “non-coherent distortion”). No big deal, but I don’t want anyone thinking it is NCM or trying to Google search that term, for it has no basis in reality. Again, no harm is done, but I want to ensure we’re using the correct terminology.
- With respect to MI Pro’s audio distortion measurement capabilities that extend well beyond any other commercial audio analyzers with the inclusion of NCD and the GedLee Metric (Gm distortion) I am very excited by their inclusion. My formal education and experience allow me to see the logic of their inclusion into a well coded audio analyzer that already has an impressive and rich feature set. I am not asking you or anyone else to use them if you (they) don’t want to. As I have repeatedly stated, “it’s a free world my friend, you and others are totally free to do as you want.” No worries if you elect to “roll your own,” (audio analyzer), etc. I’m simply trying to tell the audio community at large that that there now exist a way of making newer and more psychoacoustically relevant objective measures of audio distortion and it can be done easily and inexpensively via MI Pro.
- I’m not the only one that appreciates the newer objective ways of measuring audio distortion such an NCD and Gm distortion. Please note too that the new headphone / IEM measurement guru (Keith Howard) at the very well-known / respected website InnerFidelity will start to measure all future reviewed headphones and IEMs not only with THD and IMD but with NCD and Gm distortion as well, please see: Where we're going with Headphone Measurements | InnerFidelity. And since MI Pro is the only audio analyzer in the world with Gm distortion capabilities, you know which software he'll be using. Please also note that Keith Howard does not own an RTX6001, so my point is that MI Pro can be very useful for both RTX6001 owners as well as soundcard based audio measurement types too.
- If you read my various postings on the diyaudio website as well at Audio measurement gear | Page 9 | Audio Science Review (ASR) Forum you’ll see I have repeatedly stated that more psychoacoustic studies need to be conducted. But it is almost next to impossible to make such studies if one doesn’t have the tools available to them to start the process. That’s why I’m personally excited by these types of new objective audio distortion measurements capabilities of MI Pro. In point of fact, there may be one more added soon which is based on the work as seen in this AES paper: Perceptual Assessment of Headphone Distortion, by Louis D. Fielder, Dolby Labs, Oct 2017, (currently working out some things, but it appears to be highly likely that it too will soon be in MI Pro).
- Again, if you personally don’t want to use them, no worries my friend, don’t. I believe that there exist many in the audio community however that will start using such objective distortion measurements and conducting more research on them. That’s precisely what is needed IMO - - more well conducted psychoacoustic studies! But again, if people don’t have access to the tools to make such studies, it is pretty hard to go about making them. Hopefully you see the logic in this idea, but if not, we’ll have to simply agree to disagree as to why I see the beauty and utility in them, and why I’m advocating the audio community at large for their use.
Wishing you all the best in any event!
V/R,
MI Pro
Just to clear up some misconceptions per your recent reply to me, I provide the below thoughts:
- You made a statement that to allude to the fact that it is “my software,” which is not true in any sense of the word. As I have made clear via my 1st posting to this MI Pro GB forum and as Virtins posted in this MI Pro GB forum, I am not an employee, consultant nor in any way affiliated with Virtins in any official capacity (same with RTX Corp.)
- My main reason for getting involved with this whole MI Pro GB is that I had seen the need to have the RTX6001 unit have a custom API in order to have true autoranging and autoscaling, (just like Audio Precision gear does). To achieve this goal required I put on the hat of “coordinator / facilitator” and proposed ideas to both Virtins Technology as well as RTX Corporation. I arranged for the transfer of MI Pro USB dongle licenses to be transferred from Singapore (Virtins) to Denmark (RTX) and conversely for one RTX6001 with all mods installed and the latest firmware update applied to be shipped from Denmark to Singapore. Once this was accomplished we had to see if the “theory” of making a custom API could manifest itself into reality (an actual API capability between MI Pro and the RTX6001). I’m happy to say that prove to be the case in point.
- Once the above was done, I agreed to run a diyaudio MI Pro GB for Virtins needed a return on investment (ROI) for their time to code an API, etc. That ROI was the guaranteed sale of at least 42 MI Pro licenses. To sweeten the deal, Virtins has allowed me to offer a discounted pricing scheme on MI Pro licenses based on the total number of MI Pro licenses sold (per my posting # 1 to this GB) as well as to be able to offer a 10% discount on Virtins add-ons or the Virtins “full package” the latter which has all the MI add-on modules included.
- Before all the above happened I had pushed Virtins to include other features I wanted, (remember, I have personally purchased the MI “full package” version many years ago from them). So yes, I coordinated things between Virtins and Dr. Earl Geddes to ensure that the GedLee Metric (Gm distortion) and some other features I wanted were included into Virtins. I can only tip my hat to Virtins and give a big “thank you,” for working with me to implement things into MI I desired.
- Let’s also define the terms a bit. You continue to use the term “NCM” when you met to write “NCD” (it is called “non-coherent distortion”). No big deal, but I don’t want anyone thinking it is NCM or trying to Google search that term, for it has no basis in reality. Again, no harm is done, but I want to ensure we’re using the correct terminology.
- With respect to MI Pro’s audio distortion measurement capabilities that extend well beyond any other commercial audio analyzers with the inclusion of NCD and the GedLee Metric (Gm distortion) I am very excited by their inclusion. My formal education and experience allow me to see the logic of their inclusion into a well coded audio analyzer that already has an impressive and rich feature set. I am not asking you or anyone else to use them if you (they) don’t want to. As I have repeatedly stated, “it’s a free world my friend, you and others are totally free to do as you want.” No worries if you elect to “roll your own,” (audio analyzer), etc. I’m simply trying to tell the audio community at large that that there now exist a way of making newer and more psychoacoustically relevant objective measures of audio distortion and it can be done easily and inexpensively via MI Pro.
- I’m not the only one that appreciates the newer objective ways of measuring audio distortion such an NCD and Gm distortion. Please note too that the new headphone / IEM measurement guru (Keith Howard) at the very well-known / respected website InnerFidelity will start to measure all future reviewed headphones and IEMs not only with THD and IMD but with NCD and Gm distortion as well, please see: Where we're going with Headphone Measurements | InnerFidelity. And since MI Pro is the only audio analyzer in the world with Gm distortion capabilities, you know which software he'll be using. Please also note that Keith Howard does not own an RTX6001, so my point is that MI Pro can be very useful for both RTX6001 owners as well as soundcard based audio measurement types too.
- If you read my various postings on the diyaudio website as well at Audio measurement gear | Page 9 | Audio Science Review (ASR) Forum you’ll see I have repeatedly stated that more psychoacoustic studies need to be conducted. But it is almost next to impossible to make such studies if one doesn’t have the tools available to them to start the process. That’s why I’m personally excited by these types of new objective audio distortion measurements capabilities of MI Pro. In point of fact, there may be one more added soon which is based on the work as seen in this AES paper: Perceptual Assessment of Headphone Distortion, by Louis D. Fielder, Dolby Labs, Oct 2017, (currently working out some things, but it appears to be highly likely that it too will soon be in MI Pro).
- Again, if you personally don’t want to use them, no worries my friend, don’t. I believe that there exist many in the audio community however that will start using such objective distortion measurements and conducting more research on them. That’s precisely what is needed IMO - - more well conducted psychoacoustic studies! But again, if people don’t have access to the tools to make such studies, it is pretty hard to go about making them. Hopefully you see the logic in this idea, but if not, we’ll have to simply agree to disagree as to why I see the beauty and utility in them, and why I’m advocating the audio community at large for their use.
Wishing you all the best in any event!
V/R,
MI Pro
Let me interject my perspective.
First Gm was introduced about two decades ago. It was based on mathematical and psychoacoustically valid concepts. It was tested against THD and IMD as a measure of perception in a very narrow experiment where is was shown to have significant correlation to perception among a large group of professionally screened subjects in a double blind test. THD and IMD were shown to not have a significant correlation to perception. IMD showed a small negative correlation that was not significant.
What this means is that one can conclude that THD and IMD do not hold out any promise as valid objective metrics of nonlinear distortion. Gm does.
But, this test was only preliminary and no further tests have ever been performed. That is because based on the results of these tests I began looking elsewhere for high correlates to perception and did in fact find some. Most notably diffraction, and some other linear effects that have a nonlinear perception, making them appear to be nonlinear distortion. As a result I lost all interest in nonlinear distortion, a position that I continue to hold to this day. In loudspeakers nonlinear distortion is not a key factor in sound quality.
Is Gm the answer? I have no idea, it has potential, but so does NCD. Which is better? we simply don't know as those tests have never been done.
What we do know, and this has been verified many many times, is that THD and IMD have no correlation to sound quality no matter how one interprets them. Hence, unfortunately we really don't have a metric that has been proven to be effective. The reason IMO, is that nonlinear distortion, once the bane of audio, is now well under control and no longer a relevant factor which needs to be better resolved through improved measurements. I have found far more effective approaches looking at the other factors that I mentioned.
First Gm was introduced about two decades ago. It was based on mathematical and psychoacoustically valid concepts. It was tested against THD and IMD as a measure of perception in a very narrow experiment where is was shown to have significant correlation to perception among a large group of professionally screened subjects in a double blind test. THD and IMD were shown to not have a significant correlation to perception. IMD showed a small negative correlation that was not significant.
What this means is that one can conclude that THD and IMD do not hold out any promise as valid objective metrics of nonlinear distortion. Gm does.
But, this test was only preliminary and no further tests have ever been performed. That is because based on the results of these tests I began looking elsewhere for high correlates to perception and did in fact find some. Most notably diffraction, and some other linear effects that have a nonlinear perception, making them appear to be nonlinear distortion. As a result I lost all interest in nonlinear distortion, a position that I continue to hold to this day. In loudspeakers nonlinear distortion is not a key factor in sound quality.
Is Gm the answer? I have no idea, it has potential, but so does NCD. Which is better? we simply don't know as those tests have never been done.
What we do know, and this has been verified many many times, is that THD and IMD have no correlation to sound quality no matter how one interprets them. Hence, unfortunately we really don't have a metric that has been proven to be effective. The reason IMO, is that nonlinear distortion, once the bane of audio, is now well under control and no longer a relevant factor which needs to be better resolved through improved measurements. I have found far more effective approaches looking at the other factors that I mentioned.
That's really helpful. I hope I don't ruin my reputation here but my experience with nonlinear distortion is similar- once the distortion products are below a relatively high threshold they are not audible. Looking at the AES paper I did not see anything conclusively showing that lower distortion is essential or even identifiable. Response variations could easily sound like what we want to think some distortions sound like.
Still getting distortion out of the system is not a bad thing.
Still getting distortion out of the system is not a bad thing.
I had written a private email to EG today about this tread and I’m very happy he quickly decided to drop in to provide this comments in posting # 2303. Thank you Earl!
Taking EG’s comments into account and looking at the one well conducted study that I have provided the URLs for in my earlier posting # 2293 of NCD where it was indeed shown in a double blind AES study to correlate with how trained listeners rated sound quality of headphone playback, this simply highlights IMO the need for further studies on such matters, e.g., more with headphones, IEMs, loudspeakers, etc.
Now that an inexpensive means of measuring both NCD and Gm distortion exist, it is high time to get the tool into the hands of end users whom want to investigate such matters. For those whom don’t want to conduct such investigations, no worries, feel free to do whatever floats your boat.
I am very confident however that there exist a large segment of the audio community at large whom once the tools to make such measurements exist, that we’ll start to see more studies in this area.
I would contend that Amir’s excellent observations on this issue are spot on as well, please see Audio measurement gear | Page 9 | Audio Science Review (ASR) Forum. Does this mean that further studies in the area of audio distortion and their audibility are unwarranted? I hardly think so, for surely there exist a lot of “unknowns” that should be tested out by many researchers and end users in a vast array of audio topics, this being but one of many.
Thanks again to EG’s for taking the time to share his keen insights and thoughts on this matter!
V/R,
MI Pro
Taking EG’s comments into account and looking at the one well conducted study that I have provided the URLs for in my earlier posting # 2293 of NCD where it was indeed shown in a double blind AES study to correlate with how trained listeners rated sound quality of headphone playback, this simply highlights IMO the need for further studies on such matters, e.g., more with headphones, IEMs, loudspeakers, etc.
Now that an inexpensive means of measuring both NCD and Gm distortion exist, it is high time to get the tool into the hands of end users whom want to investigate such matters. For those whom don’t want to conduct such investigations, no worries, feel free to do whatever floats your boat.
I am very confident however that there exist a large segment of the audio community at large whom once the tools to make such measurements exist, that we’ll start to see more studies in this area.
I would contend that Amir’s excellent observations on this issue are spot on as well, please see Audio measurement gear | Page 9 | Audio Science Review (ASR) Forum. Does this mean that further studies in the area of audio distortion and their audibility are unwarranted? I hardly think so, for surely there exist a lot of “unknowns” that should be tested out by many researchers and end users in a vast array of audio topics, this being but one of many.
Thanks again to EG’s for taking the time to share his keen insights and thoughts on this matter!
V/R,
MI Pro
I guess that I may not have been clear, that my comments about nonlinear distortion were pertinent only to loudspeakers in rooms, not amplifiers or headphones. Loudspeakers have lots of audible distortion artifacts, it's just that they are far out-weighed by linear effects like FR and diffraction. Neither of these effects are present in electronics and are much smaller in headphones. Hence nonlinear distortion can be a dominant factor in these last two.
And we must remember that even though a certain measure of a particular artifact to perception might be highly correlated, that does not means that it is the most significant.
And we must remember that even though a certain measure of a particular artifact to perception might be highly correlated, that does not means that it is the most significant.
Last edited:
Hello again Earl,
Thanks for your second follow-up posting today on this nonlinear audio distortion topic. As per my prior post # 2305 where I state that I’m in complete agreement with Amir’s comments and the URL link I provided to them, it would appear that all three of us are in 100% agreement, i.e., nonlinear distortion is a real factor for headphone and in ear monitor (IEM a.k.a. “ear buds) but not so much so for loudspeakers and not the dominate factor in terms of sound quality).
Having said the above however, your comments help to explain why the aforementioned AES paper by Dr. Sean Olive and Steve Temme showing that NCD correlated well with trained listeners sound quality impressions in their double blind study correlated very well whereas THD and IMD had no correlation whatsoever. I sure wish that their study had also included Gm distortion too, (it would have been very interesting to see the results of such a study IMO).
If one looks at the historical record of all audio sells in 2016 for example, we find the following interesting comment:
“…Year to date through June, stereo headphones are the fastest-growing hi-res audio category with 64 percent U.S. dollar growth when compared to the first half of 2016...”
The above statement is admittedly a bit dated, and was taken from High Resolution Audio Device Sales See Steady Growth in the U.S. but it serves to illustrate that the younger generation and their listening / buying habits are moving the audio marketplace strongly in the direction of headphones and IEMs for audio playback. The ubiquitous nature of the cell phone market and headphone / IEM playback has resulted in a rush by the audio manufactures towards coming out with headphones and IEMs. If an audio company doesn’t get into this lucrative segment of the market, they risk losing out on a lot of potential sales.
What does the above have to do with MI Pro some may ask? Simply that I predict that the marketing departments of many audio firms in the not too distant future will start claiming that their headphone / IEMs have some ultra-low Gm distortion and/or NCD figure, (not by any means will this equate necessarily to better headphone playback automatically, but I think we’ll see some headphone and IEM companies marketing departments trying to differentiate their products from others within a very crowded marketplace).
How will such nonlinear distortion measurements be made? Since MI Pro is the only firm that has both NCD and Gm distortion measuring capabilities, at this point in time, that’s an easy question to answer
I would still like to see studies of NCD and Gm distortion on other audio products as well of course and I predict that now that an inexpensive audio analyzer incorporates these measurement models is out, it is only a matter of time until we start seeing researchers and end users testing numerous audio components in this way.
Thanks again Earl for your great comments! Looking forward to seeing a follow-up posting to the diyaudio forum from you soon (I hope) with respect to our private off-line email correspondence from earlier today.
V/R,
MI Pro
ABout ten years ago I was consulting for a major "small" loudspeaker maker in China. They were losing ground in their field and I suggested that the look at making "headphones" as it fit their processes. They lost more ground and decided to eliminate me and my ideas. I think that they threw out the baby with the bathwater. That was before Beats was bought by Apple for some ridiculous amount.
I did a major study of IEMs at the time and found some very interesting things. I thought that I had posted this study on my website. It's worth looking at if I did.
I have often pointed out that the best headphones that I had measured were the cheapest. Sold on Amazon for $35 (NOW, not what I paid years ago.) Visang R02.
I did a major study of IEMs at the time and found some very interesting things. I thought that I had posted this study on my website. It's worth looking at if I did.
I have often pointed out that the best headphones that I had measured were the cheapest. Sold on Amazon for $35 (NOW, not what I paid years ago.) Visang R02.
Last edited:
Newegg has them
ViSang VS-R02 In-Ear HiFi Noise-Isolation Earphones - Newegg.com
I had a pair and they were excellent, and I hate IEM headphones. Unfortunately a pickpocket in the Mexico DF subway thought they needed them more.
Cheers
Alan
ViSang VS-R02 In-Ear HiFi Noise-Isolation Earphones - Newegg.com
I had a pair and they were excellent, and I hate IEM headphones. Unfortunately a pickpocket in the Mexico DF subway thought they needed them more.
Cheers
Alan
I have to agree with Demian on this.
The reason that THD and IMD have little to no correlation is because it has been reduced to such ridiculously low levels that it is simply a non issue today. However There was a time where THD and particularly IMD did have a strong correlation.
Is important to look at THD and IMD, yup. Is it the end all test apparently not.
Richard, et al, has spent an entire audio career on trying put his finger on this correlation thing and he's still at it.
Major price model change to MI Pro GB for 1st 40 GB purchasers
Hot: See this diyaudio MI Pro GB posting on pricing: GB for Virtins MI Pro for RTX6001 autoranging/autoscaling & for soundcard end users - Page 8 - diyAudio
V/R
MI Pro
Hot: See this diyaudio MI Pro GB posting on pricing: GB for Virtins MI Pro for RTX6001 autoranging/autoscaling & for soundcard end users - Page 8 - diyAudio
V/R
MI Pro
Last edited:
I have made some progress regarding the USB communication with the RTX (using hidapi and Python). I now understand how to determine the button positions (and I guess it would work the same to change the settings via USB). I have tested this on Linux only, but I guess other operating systems should work too. Please send me a PM if you are interested in trying my Python test program on Windows.
Before I start implementing things in MATAA, I need to think a bit about how "autoscaling" and "autoranging" might work. I think I understand how users would use the "autoscaling" function (user sets the levels by hand using the knobs/switches, MATAA determines the button positions and automatically and scales the output and input signal data accordingly to reflect the correct voltage levels).
However, I don't quite know how "autoranging" would/should work:
Should every measurement be preceeded by an autoranging run to determine the "best" level settings? Or should autoranging only happen if the user asks for it (by executing a command)?
How would the levels be determined? For example, the autoranging could find level settings that make sure the voltages involved in a given measurement never exceed 70% of the max. value. Or are there other/better ways to determine the "best" settings?
Before I start implementing things in MATAA, I need to think a bit about how "autoscaling" and "autoranging" might work. I think I understand how users would use the "autoscaling" function (user sets the levels by hand using the knobs/switches, MATAA determines the button positions and automatically and scales the output and input signal data accordingly to reflect the correct voltage levels).
However, I don't quite know how "autoranging" would/should work:
Should every measurement be preceeded by an autoranging run to determine the "best" level settings? Or should autoranging only happen if the user asks for it (by executing a command)?
How would the levels be determined? For example, the autoranging could find level settings that make sure the voltages involved in a given measurement never exceed 70% of the max. value. Or are there other/better ways to determine the "best" settings?
If you are autoranging you change ranges when you are within some arbitrary number of dB of full scale, what I was proposing is that you allow the end user the possibility of selecting how much of the available dynamic range is used before you switch to the next range, it has nothing to do with the actual granularity of the range setting. (The range I mentioned is a better fit to the generator, for the measurement side a smaller range seems appropriate)
The challenge of autoranging in this application comes from several factors. On a traditional analyzer the autoranging has a wide range and fine steps. The Shibasoku 725 adjusts in 2.5 dB steps with a jillion relays. This is typical to keep the signal at the optimum for the system. That doesn't really work for DAC/ADC systems. First the optimum for distortion is different from optimum for SNR. Second, in the case of the RTX the steps are pretty coarse. Finer steps would be nice but quite expensive to implement and bring possible confusion.
The Liberty Praxis does address this with a digital pot in the audio interface under program control. Bill Waslo has an auto adjust routine to optimize the input signal using the pot so the display is always calibrated at at an optimum level.
My suggestion would be like this-
For level and SNR measurements switch the source at -3 dBFS up and - 23 dBFS down.
Switch the analyzer at -3 dBFS going up and -13 dBFS going down (Keep the signal no higher that - 3dBFS, except for high crest factor signals then this needs to be different. . .)
For distortion keep the source and input between -15 dBFS and -25 dBFS for the lowest distortion.
For me just reading back the actual level to scale displays would be enough value. Especially showing level and dBFS at the same time would help get the best readings.
One thing I like about DiAna is seeing the waveform, the recovered distortion waveform and the distortion spectrum at the same time. Seeing the waveform during setup helps prevent unnecessary mistakes with levels and clipping.
The Liberty Praxis does address this with a digital pot in the audio interface under program control. Bill Waslo has an auto adjust routine to optimize the input signal using the pot so the display is always calibrated at at an optimum level.
My suggestion would be like this-
For level and SNR measurements switch the source at -3 dBFS up and - 23 dBFS down.
Switch the analyzer at -3 dBFS going up and -13 dBFS going down (Keep the signal no higher that - 3dBFS, except for high crest factor signals then this needs to be different. . .)
For distortion keep the source and input between -15 dBFS and -25 dBFS for the lowest distortion.
For me just reading back the actual level to scale displays would be enough value. Especially showing level and dBFS at the same time would help get the best readings.
One thing I like about DiAna is seeing the waveform, the recovered distortion waveform and the distortion spectrum at the same time. Seeing the waveform during setup helps prevent unnecessary mistakes with levels and clipping.