This is an investigation thread into feasibility of ASR adopting the Brüel & Kjær Type 5128 "high-frequency" Head and Torso Simulator (HATS). It is on kind short-term loan from the company. As configured, it costs around US $41,000. This is quite a bit more money than competing solution. The premium comes from the 5128 benefiting from a more accurate representation of our hearing system. How much difference this makes in real life is something we need to determine in short order before I have to return the unit in four days.
Not that it matters but I like the somewhat friend/crash dummy look of the 5128
There is an external CCLD signal conditioner model 1704 which can operate on internal battery for USB. Using the latter, I saw a noise spike in upper band of the audible spectrum. Not an issue for frequency response measurements but will impact distortion data. So for that, I will be using it on battery.
One of the challenges with this fixture is that in higher frequencies it invalidates and is compatible with other measurement systems that are compliant with the so called "711" standard. Sadly, Harman target curve is not yet available either but we need something to know what the measurements mean. Without such, we would have no idea what is the property of the headphone, the measurement fixture or both.
I have been working with our resident headphone metrologist, ahd headphone technology researcher, @Mad_Economist, who has crunched a bunch of spreadsheets and models to provide two different potential target curves, the Harman one and the diffused field as provided by BK. Here it is in his words:
----------------
This file should hopefully contain everything needed to understand how this works and nothing not needed for that. To give a narrative account of how this worked, to start with I interpolated Oratory1990's Harman target data into the same divisions as the B&K data (R40 or 1/12 octave bands). I then subtracted the flat in-room baseline from the 2013 paper from the final preferred targets of the 2013, 2015, and 2018 papers to get the adjustments which were preferred on average by the listener panels.
In past, I'd have just added these to DF, since the DF-HRTF isn't that far away from the response you'd see in-room, but since I'd already done the work of making a spreadsheet with Chris Struck's methodology for in-room HRTF approximation, I decided to use that. I took the directivity data for the Revel F208 from your measurements of it, the RT60 and volume of the Harman listening room from Olive 2009, and the free and diffuse field HRTF data from the files you sent me, and plugged them into the calculator I made for Struck 2013's in-room HRTF/target inference.
As Sean has described the in-room measurement as consisting of an average of +30 degrees, -30 degrees (330 degree FF HRTF), and 0 degrees in past
I approached the in-room flat response of the 5128 two ways: first, I did an in-room response for each of the three angles and averaged them together ("Average of B+C+D" in workpage "5128 approximations of Harman"); IMO this is the closest approximation of what Todd actually did, since using three different positions gives triple weight to the indirect sound. However I additionally averaged the three free field HRTFs prior to processing through the Struck calculator to give a picture of what the flat response would be without factoring the reflections three times over ("In room flat from average of 30, 0, 330 deg FFs" in workpage "5128 approximations of Harman"). Unsurprisingly, this response was less smooth in the higher frequencies and preserved more HRTF features in fine detail.
I then applied the preference adjustments to these two "in-room baselines", and compared the results with the original Harman targets - this is in workpage "5128 Harman vs Harman". The notch like features of the 2018 adjustments strike me as odd and I have not seen any justification of them in detail from Olive, so I opted to present both the 2018/current versions and versions based on the much simpler filters in the 2015 paper in the "5128 targets" workpage.
If there's anything unclear or which you'd like elaboration on - or if you have any feedback - please let me know! I'm sort of taking this as it comes...
-------------------
I am very appreciative of his work around the clock in the last couple of days to compute this data. As he notes, the information is included in the zip file.
Note: these are draft measurements to test the fixture and for discussion. They are not formal measurements. I have performed no averaging of multiple measurements, nor playing with many headphone positions. I have a ton of headphones to measure and want to get the data out quickly. Stereo measurements are provided which should give some idea of variability although transducer differences will compound that issue.
The thread is for technical discussions only. Please don't voice objections, complaints, etc. This is a collective effort to evaluate this fixture and nothing more.
I will be measuring headphones one by one as I process them. If there is feedback on other things to do, I can include them in the ones not yet measured.
Let's see if we can figure this out everyone! Appreciate all the help in advance.
Not that it matters but I like the somewhat friend/crash dummy look of the 5128
There is an external CCLD signal conditioner model 1704 which can operate on internal battery for USB. Using the latter, I saw a noise spike in upper band of the audible spectrum. Not an issue for frequency response measurements but will impact distortion data. So for that, I will be using it on battery.
One of the challenges with this fixture is that in higher frequencies it invalidates and is compatible with other measurement systems that are compliant with the so called "711" standard. Sadly, Harman target curve is not yet available either but we need something to know what the measurements mean. Without such, we would have no idea what is the property of the headphone, the measurement fixture or both.
I have been working with our resident headphone metrologist, ahd headphone technology researcher, @Mad_Economist, who has crunched a bunch of spreadsheets and models to provide two different potential target curves, the Harman one and the diffused field as provided by BK. Here it is in his words:
----------------
This file should hopefully contain everything needed to understand how this works and nothing not needed for that. To give a narrative account of how this worked, to start with I interpolated Oratory1990's Harman target data into the same divisions as the B&K data (R40 or 1/12 octave bands). I then subtracted the flat in-room baseline from the 2013 paper from the final preferred targets of the 2013, 2015, and 2018 papers to get the adjustments which were preferred on average by the listener panels.
In past, I'd have just added these to DF, since the DF-HRTF isn't that far away from the response you'd see in-room, but since I'd already done the work of making a spreadsheet with Chris Struck's methodology for in-room HRTF approximation, I decided to use that. I took the directivity data for the Revel F208 from your measurements of it, the RT60 and volume of the Harman listening room from Olive 2009, and the free and diffuse field HRTF data from the files you sent me, and plugged them into the calculator I made for Struck 2013's in-room HRTF/target inference.
As Sean has described the in-room measurement as consisting of an average of +30 degrees, -30 degrees (330 degree FF HRTF), and 0 degrees in past
I approached the in-room flat response of the 5128 two ways: first, I did an in-room response for each of the three angles and averaged them together ("Average of B+C+D" in workpage "5128 approximations of Harman"); IMO this is the closest approximation of what Todd actually did, since using three different positions gives triple weight to the indirect sound. However I additionally averaged the three free field HRTFs prior to processing through the Struck calculator to give a picture of what the flat response would be without factoring the reflections three times over ("In room flat from average of 30, 0, 330 deg FFs" in workpage "5128 approximations of Harman"). Unsurprisingly, this response was less smooth in the higher frequencies and preserved more HRTF features in fine detail.
I then applied the preference adjustments to these two "in-room baselines", and compared the results with the original Harman targets - this is in workpage "5128 Harman vs Harman". The notch like features of the 2018 adjustments strike me as odd and I have not seen any justification of them in detail from Olive, so I opted to present both the 2018/current versions and versions based on the much simpler filters in the 2015 paper in the "5128 targets" workpage.
If there's anything unclear or which you'd like elaboration on - or if you have any feedback - please let me know! I'm sort of taking this as it comes...
-------------------
I am very appreciative of his work around the clock in the last couple of days to compute this data. As he notes, the information is included in the zip file.
Note: these are draft measurements to test the fixture and for discussion. They are not formal measurements. I have performed no averaging of multiple measurements, nor playing with many headphone positions. I have a ton of headphones to measure and want to get the data out quickly. Stereo measurements are provided which should give some idea of variability although transducer differences will compound that issue.
The thread is for technical discussions only. Please don't voice objections, complaints, etc. This is a collective effort to evaluate this fixture and nothing more.
I will be measuring headphones one by one as I process them. If there is feedback on other things to do, I can include them in the ones not yet measured.
Let's see if we can figure this out everyone! Appreciate all the help in advance.
