SRC/converter comparison

[TotalSonic]

Forumsters -
In response to the assertions made by Colin Leonard at http://www.gearslutz.com/board/6593981-post22.html that "capture with a nice AD instead of a digital SR converter it sounds better every time" - I proposed that the loopback of DAC->ADC in comparison to SRC done via a software algorithm may subjectively sound better to him in some cases but most likely is not actually closer to the sound of the original source.
I've posted a comparison test so that we can collectively examine both of our assertions.

Included is the original 24bit/88.2kHz source - first of a few seconds of -12dBFs 1kHz sine wave test tone - followed by a sine sweep going from 20Hz - 40kHz. Next is an excerpt of an in-the-box mix (not a final master) of the 2-cello string quartet Invert performing a layered improv entitled "Maroon Bells" off of our last CD "The Strange Parade" - http://www.cdbaby.com/cd/invert3

* Lavry Blue DAC (outputting at 24bit/88.2kHz) -> Apogee Soundscape 896i/o ADC (recording back in at 24bit/44.1kHZ)
* Lavry Blue DAC (outputting at 24bit/88.2kHz) -> Mytek Stereo96 ADC (recording back in at 24bit/44.1kHZ)
* RML Labs SAWStudio 4.9a software SRC (using highest quality option available)
* FMJSoft Awave Audio 10.5 software SRC (using highest quality option available)

For the analog conversions the balanced output of the DAC was plugged directly into the input of the ADC via 2' Mogami XLR cables. In both cases I attenuated the digital fader of the output of the playback workstation by 4dBFs in order to not overload the input of the ADC's.

The software SRC's reflect two different philosophies in algorithm design - in the case of SAWStudio the filter is less steep but the transition starts at a relatively lower frequency, in the case of Awave Audio the filter is steeper and the transition starts at a relatively higher frequency.

Please note that on each test file I used the gain on the Sonalkis Free G VST plugin to make sure that the 1kHz test tone at the top peaked at -12dBFs (verified to be matched down to the hundredth of a dB) on both L & R channels for all the files, and that the final fade at the end was accomplished in the DAW post capture or src (as I decided that the original length of the complete tune was just too long for the purposes of this test and didn't have time to rerun the test with the fade made on the source instead).

You can download all these files at
http://www.totalsonic.net/tsm_src_test.zip

The test files have been labelled 1, 2, 3, 4 and the order does not reflect the listing above. Comments are appreciated as to which file sounds closer to the original source (which with an src is generally the goal), and as to which file sounds subjectively "better" to you.

Folks are welcome to use the original source to convert to 24bit/44.1kHz using the converters or SRC's of your choice - you can PM me at steve at totalsonic dot net if you would like me to add these files to the comparisons (although depending on my schedule it may take me a bit to get to this).

Best regards,
Steve Berson

[Grekim]

Cool Steve. Are you going to share your personal findings on this? I tried something similar maybe 6+ years ago. It was an Apogee 8000 SE out at 48K to Apogee Rosetta in at 44.1K. This was pitted against some software converter I was using at the time, probably DP3. I don't think I took as much care as you did here, but there was no contest. The straight DAW conversion sounded best....more clarity.
I used SAW's SRC recently and was very happy with the results.
Will give your test a listen when my ear is feeling better.

[Dave Labrecque]

Forumsters -
In response to the assertions made by Colin Leonard at http://www.gearslutz.com/board/6593981-post22.html that "capture with a nice AD instead of a digital SR converter it sounds better every time" - I proposed that the loopback of DAC->ADC in comparison to SRC done via a software algorithm may subjectively sound better to him in some cases but most likely is not actually closer to the sound of the original source.
I've posted a comparison test so that we can collectively examine both of our assertions.

Included is the original 24bit/88.2kHz source - first of a few seconds of -12dBFs 1kHz sine wave test tone - followed by a sine sweep going from 20Hz - 40kHz. Next is an excerpt of an in-the-box mix (not a final master) of the 2-cello string quartet Invert performing a layered improv entitled "Maroon Bells" off of our last CD "The Strange Parade" - http://www.cdbaby.com/cd/invert3

* Lavry Blue DAC (outputting at 24bit/88.2kHz) -> Apogee Soundscape 896i/o ADC (recording back in at 24bit/44.1kHZ)
* Lavry Blue DAC (outputting at 24bit/88.2kHz) -> Mytek Stereo96 ADC (recording back in at 24bit/44.1kHZ)
* RML Labs SAWStudio 4.9a software SRC (using highest quality option available)
* FMJSoft Awave Audio 10.5 software SRC (using highest quality option available)

For the analog conversions the balanced output of the DAC was plugged directly into the input of the ADC via 2' Mogami XLR cables. In both cases I attenuated the digital fader of the output of the playback workstation by 4dBFs in order to not overload the input of the ADC's.

The software SRC's reflect two different philosophies in algorithm design - in the case of SAWStudio the filter is less steep but the transition starts at a relatively lower frequency, in the case of Awave Audio the filter is steeper and the transition starts at a relatively higher frequency.

Please note that on each test file I used the gain on the Sonalkis Free G VST plugin to make sure that the 1kHz test tone at the top peaked at -12dBFs (verified to be matched down to the hundredth of a dB) on both L & R channels for all the files, and that the final fade at the end was accomplished in the DAW post capture or src (as I decided that the original length of the complete tune was just too long for the purposes of this test and didn't have time to rerun the test with the fade made on the source instead).

You can download all these files at
http://www.totalsonic.net/tsm_src_test.zip

The test files have been labelled 1, 2, 3, 4 and the order does not reflect the listing above. Comments are appreciated as to which file sounds closer to the original source (which with an src is generally the goal), and as to which file sounds subjectively "better" to you.

Folks are welcome to use the original source to convert to 24bit/44.1kHz using the converters or SRC's of your choice - you can PM me at steve at totalsonic dot net if you would like me to add these files to the comparisons (although depending on my schedule it may take me a bit to get to this).

Best regards,
Steve Berson

Steve -- it may be too late, but did you consider making a second, randomized set of the four files that could be used for the "sounds better" determination, so that the trial "blindness" would be complete across the board? There could be some "placebo effect" influencing that judgement on the heels of the "closest to the source" judgement (or vice-versa, depending on the order of the two assessments).

[TotalSonic]

Steve -- it may be too late, but did you consider making a second, randomized set of the four files that could be used for the "sounds better" determination, so that the trial "blindness" would be complete across the board? There could be some "placebo effect" influencing that judgement on the heels of the "closest to the source" judgement (or vice-versa, depending on the order of the two assessments).

Dave -
Sounds like what you are wanting is actually "ABX" type testing software - which if you do a net search you can find a good number of options for Windows readily available. Another way is simply to have someone else playback the files to you in random order between playing back the original source.

Anyway - unfortunately I don't have any time in the next week or so to refine this test myself - but folks are welcome to use the source to help construct their own further tests if they wish.

Best regards,
Steve Berson

[Ian Alexander]

Steve,

Can you offer more detail on the reason for the -4dB change on the analog tests? You mentioned avoiding overloading the ADC input, so does this mean there are some overs in the output?

Would adjusting the analog method by 4dB affect the sound? Certainly a 4dB drop might be perceived as lower quality - at least pushy audio store salesmen think that consumers feel that louder is better. But even if the level was adjusted back up 4dB after the ADC, would the additional math done by the lowering and raising make the playing field unequal?

I'm sure you realize, but others might not, that having someone play the files in random order doesn't qualify as the best type of test - double blind - which requires that neither the participant nor the administrator know which is which. I'm guessing the software you mentioned would eliminate the admin bias.

Thanks for starting the discussion and providing the files.

[TotalSonic]

Steve,

Can you offer more detail on the reason for the -4dB change on the analog tests? You mentioned avoiding overloading the ADC input, so does this mean there are some overs in the output?

No - the output was completely clean and absolutely free of any overs. The reason it is lowered is so that the input of the Apogee ADC (which does not have an attenuator built into it) would not be overloaded itself.



Would adjusting the analog method by 4dB affect the sound? Certainly a 4dB drop might be perceived as lower quality - at least pushy audio store salesmen think that consumers feel that louder is better. But even if the level was adjusted back up 4dB after the ADC, would the additional math done by the lowering and raising make the playing field unequal?

To my ear the attenuation of 4dB using SAWStudio's fader is essentially an inaudible change. You can test this for yourself by lowering the fader - processing the file and then raising the fader backup and processing the file - and then do a null test against the original source (line up two files to sample level accuracy separate inputs and invert the polarity of the second track and render the results to a new file) to see what the differences between them is.



I'm sure you realize, but others might not, that having someone play the files in random order doesn't qualify as the best type of test - double blind - which requires that neither the participant nor the administrator know which is which. I'm guessing the software you mentioned would eliminate the admin bias.

I'm hoping that those who are truly interested in making this a scientifically accurate and unbiased test will be aware of the methods needed to achieve this.



Thanks for starting the discussion and providing the files.

you're welcome!

Best regards,
Steve Berson

[Carl G.]

Steve,
I've PM'd my results.

[TotalSonic]

Steve,
I've PM'd my results.

Carl -
Please post them here! It will be easier to discuss them amongst each other that way. Thanks for taking time to listen and for posting your evaluations!

Best regards,
Steve Berson

[DominicPerry]

Once again my magic ears can detect no difference between any of them.

It has to be said that something must be SRC'ing the original for me, because the interface I'm using doesn't support 88.2KHz.

Dominic

[Dave Labrecque]

Once again my magic ears can detect no difference between any of them.

LOL. :p

[TotalSonic]

Once again my magic ears can detect no difference between any of them.

Dominic -
Honestly there are indeed definitely audible differences between the comparison files! Maybe you should look into upgrading your monitors and room if you have the desire and budget. Otherwise - ignorance is bliss! ;)



It has to be said that something must be SRC'ing the original for me, because the interface I'm using doesn't support 88.2KHz.

What sound card are you using? Many of them (including my Sydec Mixtreme 192) are capable of doing SRC on the fly - and sometimes these thngs can be turned on by default.

Best regards,
Steve Berson

[DominicPerry]

Dominic -
Honestly there are indeed definitely audible differences between the comparison files! Maybe you should look into upgrading your monitors and room if you have the desire and budget. Otherwise - ignorance is bliss! ;)

What sound card are you using? Many of them (including my Sydec Mixtreme 192) are capable of doing SRC on the fly - and sometimes these thngs can be turned on by default.



Since I sold my Genelec 8040's I've been listening (rather than monitoring) on a Cyrus Amp and Mission 780 speakers. Far from ideal. I think any kind of real monitor would be better. Also, I was using Foobar to play back (although I have just re-acquired SAW, and would normally load the files into that and hot-track solo between them to compare). I can't find any SRC settings in Foobar. I'm also using Win7 64bit, and I can't find the SRC settings in that either, but they're probably around. The interface is a SoundDevices USBPre, running an third party ASIO driver, as no 64bit driver is available from SoundDevices. The interface is 24bit/48/44.1KHz in and 16bit/48/44.1KHz out. But the drivers are 24 bit and the hardware chops the output (and may or may not dither). Very odd. But it's tough and compact with reasonable micpres.
It's not a problem, I'm not too fussy about these things, I just find it amusing that I can almost never hear the differences engineers worry about. :D

EDIT: I put the files in SAW. Of course the original is being SRC'd by SAW so I'm guessing that should make it the same as one of the other four. But of course, I can't tell which one, 'cos they all still sound the same!
On the other hand, having listened to this clip quite a few times, I'm beginning to rather like it - reminds me of a sparse version of Schoenberg's Verkl***228;rte Nacht. I might order a copy of the CD. Unless I can get a 32kb/s mp3 ;)


Dominic

[Carl G.]

Carl -
Please post them here! It will be easier to discuss them amongst each other that way. Thanks for taking time to listen and for posting your evaluations!

Best regards,
Steve Berson

Steve,
1 & 4 are similar and most like original (testing specifically 1:40-1:45)
2 & 3 are similar and to me sound more openly dynamic than original (I prefer #2 and #3, but if I need completely accuracy #1 & #4 would be the choice)
(testing specifically 1:40-1:45)
I'm listening closely through MDR-V6 each through SoundFile in SS.
After the listening tests, I've closely examined wave data - which confirms my findings of similarities above.
(sorry for the delay - working on power sources - computer been off)

[Gary Ray]

Steve,
1 & 4 are similar and most like original (testing specifically 1:40-1:45)
2 & 3 are similar and to me sound more openly dynamic than original (I prefer #2 and #3, but if I need completely accuracy #1 & #4 would be the choice)
(testing specifically 1:40-1:45)
I'm listening closely through MDR-V6 each through SoundFile in SS.
After the listening tests, I've closely examined wave data - which confirms my findings of similarities above.
(sorry for the delay - working on power sources - computer been off)

I used a scale of 1-5 for correlation to the original, 5 being very much like the original.
I agree on the findings of 1 and 4 are similar and most like the original. I gave them both a correlation number of 4.
I gave 2 and 3 a correlation number of 2 for both. I did not find any differences in dynamics between them (1-4).

I loaded everything on one of my Sound Devices 722. I then listened during separate sessions with AKG 701 and Sennheiser 800 cans. I used Sound Forge to look at the wave forms and agree that it is easy to see differences.

Steve, quite an interesting test.
Thanks,

[Bill Park]

Dominic -
What sound card are you using? Many of them (including my Sydec Mixtreme 192) are capable of doing SRC on the fly - and sometimes these thngs can be turned on by default.

Best regards,
Steve Berson

When I was looking for my newest sound card, someone promoted the Lynx AES card because in one of it's iterations it performs SRC on the fly, supposedly pretty well. (In the end I went with the RME AES card.)

[Naturally Digital]

When I was looking for my newest sound card, someone promoted the Lynx AES card because in one of it's iterations it performs SRC on the fly, supposedly pretty well. (In the end I went with the RME AES card.)There's a Lynx AES16/SRC version. It does the SRC in hardware. Also available on the AES input of the LynxTwo and L22.

[Ian Alexander]

I'm sure you realize, but others might not, that having someone play the files in random order doesn't qualify as the best type of test - double blind - which requires that neither the participant nor the administrator know which is which. I'm guessing the software you mentioned would eliminate the admin bias.

An apology is in order here. I did not read the OP carefully enough to notice that you had already changed the order and renamed the files. I'm sorry, Steve.:o