In our never-ending quest for the ultimate sound quality, we have been looking for tools and techniques that help us get precisely there. And today, let me tell you the story of how we ended up changing our entire production chain to provide you with better quality music whilst lowering our overall resolution.
From 24bit to 32bit floating point
Ever since the very beginnings of TRPTK and up until a relatively little while ago, we have recorded all our albums in DXD 352.8kHz 24bit, and mastered in 32bit floating point. About a year and a half ago, we first noticed that the 32bit floating point output of the digital audio workstation we used back then (Merging Technologies’ Pyramix, the software pretty much everyone in our industry uses) would sound dramatically better than its 24bit output. Strange, we thought, since 24 bits is more than plenty in terms of dynamic range (144dB to be precise), and the difference was so big we couldn’t attribute it to just the word length or dynamic range itself. Something else had to be at play here. We did a double-blind test, where we rendered a master at both 24bit and 32bit floating point, and played it back through our software of choice, Roon. Our entire team would be able to hear the differences, almost every time preferring the 32bit floating point version.
Mind you, the only thing that should differ between these files is that the 24bit version would have some kind of dithering/noise-shaping applied to account for the word length reduction. And so we did another test. By setting up our streamer in a way that it automatically reduces/dithers/noise-shapes the 32bit floating point version to 24bit, we were now comparing two different 24bit versions: one created by our mastering software, and one created by our streaming software. Guess which won? That’s right, Roon did. And by a fair margin.
Since we now figured that our streaming software did a better job at dithering/noise-shaping than our mastering software, perhaps there was software that could do this even better! After a long period of trying out all different kinds of software, we found that Weiss Engineering Saracon did just that. Comparing a 24bit DXD file rendered from the same 32bit source as the 24bit file created in Pyramix was mind-boggling. We didn’t expect such a huge difference. Such a huge difference indeed, that we decided to “upgrade” our entire catalogue, by going back to all the original sessions, rendering the raw 32bit floating point output, and then creating our downloadables through Saracon. (This process is now almost complete, by the way, so most downloads in our shop are the “upgraded” ones.)
We also found out that when you apply any kind of DSP (room correction, sample rate conversion or whatever else) through your streaming software, it would be better to do so from the original raw 32bit floating point output rather than the 24bit version, so we began offering this 32bit downloadable too. And it seemed like our journey would end there.
From DXD to DSD256
Impressed by the sheer conversion quality of Saracon, we were wondering whether we could harness said quality to record in one format, and edit/mix/master in another. For example, our ADCs (the Merging Technologies Hapi outfitted with 16 channels of DXD/DSD256), would run at DSD256 whilst recording, and then when it’s time for editing, mixing and mastering (which cannot be done in DSD), we would convert all recorded media to some kind of PCM format (352.8kHz at 32bit or 64bit floating point). This also yielded vastly improved sound quality compared to recording in DXD, in terms of their richness of detail and impulse response! We were delighted when we found out, and once again changed our recording and mastering chain accordingly. All in the name of superb sound quality!
Going further, going back
Of course, my inquisitive mind was wondering how far we could take this. Perhaps we could record in DSD256 and then edit/mix/master at an extraordinarily high sample rate and bit depth so that this entire step in the process would become sonically 100% transparent? We purchased a DAC that was able to handle such resolutions and started experimenting with post-production (editing/mixing/mastering) in 705.6kHz 64bit floating point. For this, I needed to change my mastering software as Pyramix can only go up to 352.8kHz 32bit floating point. I chose Cockos Reaper, a partly-open-source piece of software that can do just about everything save for brewing coffee. And again, we heard some improvements over our then-current software. Fantastic! Yet again, we changed the entire mastering chain to include this step, all in the name of good sound quality.
As you might understand, we changed three variables: the sample rate (doubling from 352.8kHz to 705.6kHz), the bit depth of all processing involved as well as the output files (doubling from 32bit floating point to 64bit floating point), and the software in which we did post-production (from Pyramix to Reaper). Of course, I was wondering which of these three made the biggest impact in sound quality, and so I created another experiment in which I would create a whole bunch of different masters, with all combinations of the following variables:
Sample rate (352.8kHz and 705.6kHz)
Bit depth of processing (32bit floating point and 64bit floating point)
Bit depth of output file (32bit floating point and 64bit floating point)
Software used (Pyramix, Reaper and Steinberg Nuendo)
After extensive listening tests, I found I pretty much always preferred either the Reaper or Nuendo files, but weirdly enough, that I would also prefer the 352.8kHz versions over the 705.6kHz versions in a lot of cases. So what gives?
Well, this all has to do with the processing applied to our masters. Of course, we like to apply as little processing as we can, but in some cases you just simply have to do something. For example, in projects involving more than just the main set of microphones, you would want to use a bit of EQ on the added close-mics to filter away low frequency rumble in order to keep clarity in your mix. Or, on extremely dynamic projects, you might need to apply some extremely subtle limiting in order for peaks not to go into distortion, or to keep everything at a listenable level. We noticed that each time we applied any of this processing, the 705.6kHz versions would sound worse than they did in 352.8kHz, which to me sounds like one clear thing: these processing plugins cannot handle 705.6kHz and thus apply sample-rate conversion on the input and output. By keeping at 352.8kHz we were able to use all plugins we would need at their maximum allowed resolution without them resampling the beauty out of them.
So why am I telling you this long and boring story?
Well, first of all, we do this to explain why we have changed our entire production chain yet again, but this time for the very long term. We’ve been through numerous resolutions, sample rates, bit depths, software packages, plugins, and processing chains, but finally we ended up where we are right now.
As it stands, our production chain looks like this: we first record every project in DSD256, because that’s the resolution our analogue-to-digital-converters run best on. We then convert all these DSD256 files we just recorded to PCM 352.8kHz 64bit floating point, to be able to work in a format that’s as close to sonically transparent as absolutely possible. These files are then brought into Nuendo, a piece of high-end post-production software, where we do all our editing and mixing. We then bring the output of this into another piece of software called WaveLab (made by the same people who created Nuendo), where we master the final product. We export its raw 352.8kHz 64bit floating point output into a file, and use Weiss Saracon to create all our downloadable albums you can find on our website.
So what does this mean to you, our dear listeners? For starters, it means that the couple of 705.6kHz albums we had on offer will be replaced by 352.8kHz versions that, almost unintuitively, sound better than their 705.6kHz counterparts. It also means that, for those who want to apply their own DSP or sample rate conversion, our raw 64bit floating point downloads will be available, which might offer slightly enhanced sound quality when used in conjunction with DSP. Your mileage, however, may vary of course. The 705.6kHz versions might over time disappear from our website, in favour of better sounding 352.8kHz counterparts.
Which format should I download then?
In almost all cases, we recommend the 352.8kHz 24bit FLAC file. It’s relatively compact, and since it’s mastered at 64bit and then converted using Saracon the sound quality is utterly amazing. If you’re adventurous and would like to experiment with adding your own DSP, sample rate conversion, dithering or noise shaping, you can check out the 64bit floating point ones, since that will save you at least one conversion step in the process. It’s actually that simple.
In the end, we don’t want to offer you any big numbers. We don’t care about head-spinning, mind-blowing numbers, we care about the ultimate sound quality. And we hope you do too.
Lower Resolution, Better Sound Quality?
In our never-ending quest for the ultimate sound quality, we have been looking for tools and techniques that help us get precisely there. And today, let me tell you the story of how we ended up changing our entire production chain to provide you with better quality music whilst lowering our overall resolution.
From 24bit to 32bit floating point
Ever since the very beginnings of TRPTK and up until a relatively little while ago, we have recorded all our albums in DXD 352.8kHz 24bit, and mastered in 32bit floating point. About a year and a half ago, we first noticed that the 32bit floating point output of the digital audio workstation we used back then (Merging Technologies’ Pyramix, the software pretty much everyone in our industry uses) would sound dramatically better than its 24bit output. Strange, we thought, since 24 bits is more than plenty in terms of dynamic range (144dB to be precise), and the difference was so big we couldn’t attribute it to just the word length or dynamic range itself. Something else had to be at play here. We did a double-blind test, where we rendered a master at both 24bit and 32bit floating point, and played it back through our software of choice, Roon. Our entire team would be able to hear the differences, almost every time preferring the 32bit floating point version.
Mind you, the only thing that should differ between these files is that the 24bit version would have some kind of dithering/noise-shaping applied to account for the word length reduction. And so we did another test. By setting up our streamer in a way that it automatically reduces/dithers/noise-shapes the 32bit floating point version to 24bit, we were now comparing two different 24bit versions: one created by our mastering software, and one created by our streaming software. Guess which won? That’s right, Roon did. And by a fair margin.
Since we now figured that our streaming software did a better job at dithering/noise-shaping than our mastering software, perhaps there was software that could do this even better! After a long period of trying out all different kinds of software, we found that Weiss Engineering Saracon did just that. Comparing a 24bit DXD file rendered from the same 32bit source as the 24bit file created in Pyramix was mind-boggling. We didn’t expect such a huge difference. Such a huge difference indeed, that we decided to “upgrade” our entire catalogue, by going back to all the original sessions, rendering the raw 32bit floating point output, and then creating our downloadables through Saracon. (This process is now almost complete, by the way, so most downloads in our shop are the “upgraded” ones.)
We also found out that when you apply any kind of DSP (room correction, sample rate conversion or whatever else) through your streaming software, it would be better to do so from the original raw 32bit floating point output rather than the 24bit version, so we began offering this 32bit downloadable too. And it seemed like our journey would end there.
From DXD to DSD256
Impressed by the sheer conversion quality of Saracon, we were wondering whether we could harness said quality to record in one format, and edit/mix/master in another. For example, our ADCs (the Merging Technologies Hapi outfitted with 16 channels of DXD/DSD256), would run at DSD256 whilst recording, and then when it’s time for editing, mixing and mastering (which cannot be done in DSD), we would convert all recorded media to some kind of PCM format (352.8kHz at 32bit or 64bit floating point). This also yielded vastly improved sound quality compared to recording in DXD, in terms of their richness of detail and impulse response! We were delighted when we found out, and once again changed our recording and mastering chain accordingly. All in the name of superb sound quality!
Going further, going back
Of course, my inquisitive mind was wondering how far we could take this. Perhaps we could record in DSD256 and then edit/mix/master at an extraordinarily high sample rate and bit depth so that this entire step in the process would become sonically 100% transparent? We purchased a DAC that was able to handle such resolutions and started experimenting with post-production (editing/mixing/mastering) in 705.6kHz 64bit floating point. For this, I needed to change my mastering software as Pyramix can only go up to 352.8kHz 32bit floating point. I chose Cockos Reaper, a partly-open-source piece of software that can do just about everything save for brewing coffee. And again, we heard some improvements over our then-current software. Fantastic! Yet again, we changed the entire mastering chain to include this step, all in the name of good sound quality.
As you might understand, we changed three variables: the sample rate (doubling from 352.8kHz to 705.6kHz), the bit depth of all processing involved as well as the output files (doubling from 32bit floating point to 64bit floating point), and the software in which we did post-production (from Pyramix to Reaper). Of course, I was wondering which of these three made the biggest impact in sound quality, and so I created another experiment in which I would create a whole bunch of different masters, with all combinations of the following variables:
After extensive listening tests, I found I pretty much always preferred either the Reaper or Nuendo files, but weirdly enough, that I would also prefer the 352.8kHz versions over the 705.6kHz versions in a lot of cases. So what gives?
Well, this all has to do with the processing applied to our masters. Of course, we like to apply as little processing as we can, but in some cases you just simply have to do something. For example, in projects involving more than just the main set of microphones, you would want to use a bit of EQ on the added close-mics to filter away low frequency rumble in order to keep clarity in your mix. Or, on extremely dynamic projects, you might need to apply some extremely subtle limiting in order for peaks not to go into distortion, or to keep everything at a listenable level. We noticed that each time we applied any of this processing, the 705.6kHz versions would sound worse than they did in 352.8kHz, which to me sounds like one clear thing: these processing plugins cannot handle 705.6kHz and thus apply sample-rate conversion on the input and output. By keeping at 352.8kHz we were able to use all plugins we would need at their maximum allowed resolution without them resampling the beauty out of them.
So why am I telling you this long and boring story?
Well, first of all, we do this to explain why we have changed our entire production chain yet again, but this time for the very long term. We’ve been through numerous resolutions, sample rates, bit depths, software packages, plugins, and processing chains, but finally we ended up where we are right now.
As it stands, our production chain looks like this: we first record every project in DSD256, because that’s the resolution our analogue-to-digital-converters run best on. We then convert all these DSD256 files we just recorded to PCM 352.8kHz 64bit floating point, to be able to work in a format that’s as close to sonically transparent as absolutely possible. These files are then brought into Nuendo, a piece of high-end post-production software, where we do all our editing and mixing. We then bring the output of this into another piece of software called WaveLab (made by the same people who created Nuendo), where we master the final product. We export its raw 352.8kHz 64bit floating point output into a file, and use Weiss Saracon to create all our downloadable albums you can find on our website.
So what does this mean to you, our dear listeners? For starters, it means that the couple of 705.6kHz albums we had on offer will be replaced by 352.8kHz versions that, almost unintuitively, sound better than their 705.6kHz counterparts. It also means that, for those who want to apply their own DSP or sample rate conversion, our raw 64bit floating point downloads will be available, which might offer slightly enhanced sound quality when used in conjunction with DSP. Your mileage, however, may vary of course. The 705.6kHz versions might over time disappear from our website, in favour of better sounding 352.8kHz counterparts.
Which format should I download then?
In almost all cases, we recommend the 352.8kHz 24bit FLAC file. It’s relatively compact, and since it’s mastered at 64bit and then converted using Saracon the sound quality is utterly amazing. If you’re adventurous and would like to experiment with adding your own DSP, sample rate conversion, dithering or noise shaping, you can check out the 64bit floating point ones, since that will save you at least one conversion step in the process. It’s actually that simple.
In the end, we don’t want to offer you any big numbers. We don’t care about head-spinning, mind-blowing numbers, we care about the ultimate sound quality. And we hope you do too.