 |
| Impulse Response: One of the talking points from back in the day as a selling point for DSD... Yup! DSD can better reproduce a 0.000003 second "click". Source: Merging Technologies |
I. Preamble
It is amazing how quickly another year has passed. About this time last year, I posted the first comparison of DSD Encoders and Decoders "shoot-out" of sorts comparing Weiss Saracon 01.61-27, KORG AudioGate 2.3.3 and JRiver 19.0.117 in terms of quality - both encoding and decoding fidelity using the RightMark Audio Analyzer software. The idea was to determine which of the three created DSD files from an original 24/96 PCM test signal and then decoded it back to 24/96 in a way where there was as little change in terms of distortion, flat frequency response, and lowest amount of added noise.Perhaps not surprisingly, the expensive Weiss Saracon software sets the standard as the most consistent DSD encoder that resulted in the best output once decoded. The differences in decoding capability appeared to be very minor (questionable audibility between the 3) but objectively, both Saracon and JRiver 19 were on par and the free AudioGate 2 somewhat "noisier" in terms of the PCM output (I speculated this was due to stronger dithering algorithm).
Well, another year has passed in terms of software upgrades to DSD decoding and I was interested to compare the decoding capabilities as of late. We have brand new versions of JRiver and AudioGate now, plus I didn't get to test foobar with the DSDIFF plugin last year. Plus we now have DSD decoding on the Mac OS X available with XLD and commercially with DSD Master.
To maintain an "apples to apples" comparison as best I can, I will use the Saracon 01.61-27 encoded DSD file of a 24/96 test signal with each of these decoders. I think this is fair given that Saracon produced excellent results last time as an encoder plus it is an "industry standard" used in many professional studios.
Let us have a look at the decoders I will be testing:
1. foobar with DSDIFF 1.4 plugin [Windows]. foobar2000 does not need any introduction as a freely available Windows music player. It's stable. Sounds fantastic. Is bitperfect. Is extremely feature-rich. And as witnessed by the myriad plugins like DSDIFF, very extensible. The DSDIFF plugin has been around as version 1.4 since 2011 by kode54. Decoding was set to 24/96 PCM, WAV output.
2. KORG AudioGate 3.0.2 [Windows]. AudioGate got an upgrade in June 2014 to version 3. It's no longer "free" like before where you can do conversions after just allowing Tweets from your account. I'm going to test with the output set at 24/96, "High Quality", no dithering. Let's see if the noise floor is better than AudioGate 2.
3. JRiver Media Center 20.0.87 [Windows used, Mac and Linux also available]. Upgrade from version 19 to 20 has introduced some new features (one of which I'm most interested in I'll talk about another time). Don't know if DSD decoding has changed... We shall see! For the test this year, I decided to stay with the default "safe" 24kHz low-pass filter like last year. Remember, you do have a choice: go to Tools --> Options... --> Advanced --> ... Configure input plug-in ... --> DSD input plug-in...
4. X Lossless Decoder (XLD) 20141129 [Mac OS X]. Freely available audio tool for Mac users. As of the 2014/11/09 release, it has the capability of decoding DSD files. Let's see how this newcomer compares! Again, decoding target was 24/96, I used higher quality "SoX VHQ Linear" resampling. Since the options allow easy adjustment of parameters like decimation and quantization, I will test both the default (8:1 decimation, 24-bit integer quantization) as well as higher quality (8:1 decimation, 32-bit floating point). [Decimation correlates to the output samplerate, so 8:1 for DSD64 is 352.8kHz output which then gets resampled by SoX to 96kHz.]
5. DSD Master 1.0 [Mac OS X]. Thank you to Richard at BitPerfect Sound Inc. for letting me give this program a spin (US$29.99 on the AppStore)! As per the company namesake, these are the same guys who brought you BitPerfect for the Mac. I actually downloaded this program in late 2014, unfortunately it took me awhile to get to the testing... After migrating to Windows in the last 2 years, I just haven't been using the Mac nearly as much. As with the other programs, I've set DSD Master to do all processing back to 24/96. Write as a WAV file. No gain applied. I see DSD Master can handle up to DSD256. One interesting feature is the DSD Hybrid mode where the DSD data is retained along with PCM conversion - large file sizes to be expected, and you will need BitPerfect 2.0 to play through iTunes to a DSD DAC.
For completeness, I will also post up the results for Weiss Saracon 01.61-27 which I obtained last year as the standard for comparison. (I borrowed Saracon to test last year so have not kept up as to whether there have been updated versions since.)
II. Results
Here are the summary results (to keep the tables smaller, I separated the Windows and Mac software used): |
| Windows DSD Conversion |
 |
| Mac DSD Conversion |
If we compare the software updates, we see that there has been a substantial change with KORG AudioGate 3 compared to version 2 in terms of noise level; I'm seeing about a 10dB improvement compared to last year. If my suspicion is correct, perhaps they were using a stronger dithering algorithm back in version 2. The result is now much more in line with the other conversion packages.
As for JRiver 20 vs. 19, there has been little change; about 1-2dB difference. Note that JRiver posted fantastic numbers last year already so the slight improvement is actually very impressive!
Let's now look at the newcomers to this round-up...
Foobar DSDIFF did okay overall. Good performance and about the same as AudioGate when decoding the test signal. Hey, it's free :-).
XLD on the Mac provided good results as well. As you can see comparing the 24-bit integer to 32-bit floating point calculations, there was no difference. Any difference between 24-bit and 32-bit processing would be below the precision of the final 24/96 WAV file and RightMark's calculations. Therefore, if you use this program, you might as well stick with 24-bit integer calculations as it's faster.
Finally, we have the DSD Master software. Excellent quality output converting DSD64 to 24/96. Very low noise level and high dynamic range essentially the same as the much more expensive Weiss Saracon. (Remember, this is all relative since we're talking about noise levels below -130dB!)
Let's have a look at the graphs:
 |
| Windows DSD Conversion - Frequency Response |
 |
| Mac DSD Conversion - Frequency Response |
On the Mac, I see that neither XLD nor DSD Master perform much low-pass filtering - at least not within 48kHz.
Here are the noise level graphs:
 |
| Windows DSD Conversion - Noise Level. Slight irregularity in the DSDIFF noise floor at high frequencies. |
 |
| Mac DSD Conversion - Noise Level [note XLD tracings exactly the same so only the purple 32-bit tracing showing up] |
III. Conclusions
As I suggested last year, I believe that DSD --> PCM conversion is transparent. I'm measuring the distortion added by both PCM (24/96) --> DSD64 [via Saracon] as well as DSD64 --> PCM (24/96) steps and as you can see, there is nothing showing up of concern. Fidelity is maintained beyond any DAC's analogue output I am aware of except for all that ultrasonic stuff peaking at about -85dB if filtering is not applied. Speaker system / headphone playback would add more distortion than this (at least within the audible frequencies).You might be asking - what about subjective listening?
Bottom line: No need to worry about the sonic output from any of these converters IMO. Conversion algorithms and software look mature with little difference between them. The only significant choice is whether you want to have a low-pass filter in the conversion process (I do so I'd prefer Saracon or JRiver). I know some people claim they can hear qualitative differences between conversion programs beyond just level differences... Maybe. I'd certainly be impressed if anyone can show positive controlled, blinded listening test results given the minute changes I see/hear while doing these tests!
-------------------
So guys, what do you think about the state of DSD these days? It really looks like the "push" has fizzled lately... Other than more DACs supporting native DSD playback, there seems to be little news out there. Anyone actually buying many DSD downloads?
As I wrote back in April 2013 (On SACD & DSD audio...), there are many factors working against DSD audio if the goal is to expand beyond just a small audio-geek niche format. It appears my concerns around the need for a modern file format that provides full tagging and data compression persists...
Psssstttt... Coders... Want to be famous? Pull together some code to create an open source ID3 taggable compression CODEC for DSD (.fdac? Free DSD Audio Codec - how about just .dac format). Get the guys at JRiver and foobar2000 to support it, and make sure it runs in Linux for music servers. I bet this format would become widely used among the guys ripping their SACD's and those who rip LPs into DSD! Make sure to support compression of DSD128+ as well just in case hi-res DSD becomes available (as far as I know the old Philips ProTECH DST Encoder could only handle DSD64)...
Until next time, have a wonderful April and hope you're enjoying the music!
Interesting test..
ReplyDeleteKuddos for the great write-up.
I could never resist a smile when someone pulls out the opening picture in this post as a comparative picture to show superiority of DSD.
One should be aware that the 3us pulse is half the period of a 166.7 kHz squarewave at 0dB.
It has an almost infinite frequency response and needs amplification bandwidth of at least >1MHz.
ONLY a squarewave generator could generate such a pulse and there is no analog audio equipment (be it reel to reel nor vinyl cutter/needle/pre-amp that could ever reproduce such a pulse.
For analog tape equipment the pulse would probably look something like the 96kHz (without pre-ringing) in the very best reel-to-reel equipment.
NO vinyl rig could ever reproduce this pulse.
Given the fact that there aren’t many transducers (speakers/headphones) around that can accurately reproduce frequencies above 50kHz around it puts the example signal in another perspective.
Aside from this there is NO such signal 'pulse' present in any music signal at all rendering the picture completely pointless.
Also there is NO audio signal not music signal that has any rise-/fall-times nor transients that will EVER come near the 3us pulse in the picture.
All recorded signals (if the mics could pick it up) will go through a low pass (brickwall) filter BEFORE entering the ADC in the studio.
I am not aware of recording equipment with bandwidths >200kHz which would be needed to record such a pulse.
B.T.W.
I have no experience with DSD and can't/won't comment on the sound.
My reply is only about the first picture.... which one sees everywhere and is complete and utter nonsense.
P.S.
DeleteThe DSDIFF plugin is not that current any more and has basically been replaced by the SACD plugin.
It may be interesting to try it out and measure it ?
Also testing HQ player with its myriad of filters might be interesting.
Thanks Solderdude. I'll have a look at SACD plugin since the testing is still "fresh" :-)
DeleteA few other thoughts have come to mind so I'll put up a full post next week looking at this as well as a few other nuggets to consider!
Oh yes Frans, I almost forgot to comment on the impulse response discussion...
DeleteGet your fingers and keyboard warmed up :-). Probably in the not-too-distant future, there will be plenty of opportunities to comment on the "virtues" of impulse responses. I sense the forces of the "Microsecond Click Brigade" amassing under the banner of "Meridian MQA".
Let's see how this plays out...
Yes, I can see that happening.
DeleteResearch has shown that people can actually 'detect' risetimes faster than 20kHz risetimes under lab conditions with test tones, no definitive proof with more complex music signals yet, although the 192kHz+ and DSD brigade will think otherwise based on what they and/or noted (fill in guru of choice) has shown without a shadow of a doubt.
Perhaps responders would like to share links to properly performed research showing >20kHz BW is needed for music and or pictures of music signals (at bit level) that actually show extremely fast risetimes in music.
What I find amusing is the 'rustling leaves pin point accuracy' which is then coupled to high frequency detection and risetimes so one knows where to run to when a predator sneakes up on us. This would be the 'evolutionary' proof that we are build to do this for our survival. It would allow us to do more accurate imaging for music.
I have always wondered about this as I played with funny circuits that you can hide in a room and either emit a 'cricket' (a short burst of needle pulses) or a short high pitched 'beep' (like a smokedetector does when the battery is about to become dead).
Both are very annoying sounds that are almost impossible to locate by ear. It takes a lot of time to find these 'bugs' in a room, due to reflections I guess. Never tried it outdoors though.
So much for pin point accuracy of hearing and needle pulses in rooms (let's even forget speakers).
On headphones is easier to pinpoint a place between L and R. Some seem to hear sounds in front of the rear as well (funny brains ?) I only hear L and R and inbetween but not everyones brain works the same.
Ears should be used to enjoy music with, not analyse equipment.
"Research has shown that people can actually 'detect' risetimes faster than 20kHz risetimes under lab conditions with test tones"
DeleteCould you link those studies here? I'm sure many of would be interested in having a read!
Good work!
ReplyDeleteLike i said other time, i think that what we need is high quality recordings in multichannel. This will make difference. But it has to be DSD? PCM? I don't know and really don't care (although i'm more inclined to PCM, since it is a mature and highly supported tech...).
I think that the future of audio (and video) should be software based, and not so much hardware based, and if engineers and programmers do their job well done, we can acheive astonishing results (just look at the outstanding audio quality that one can have these days only with a notebook and a headphone at a mid-budget price!!!).
Right now i don't have the $$$ for entering the hi-fi multichannel audio world, but i'm looking foward to multi-ch, even from 2 CH recordings, using proper and accurate DSP. Let's see...
As always, keep up the good work! One can feel at home in here!
Best regards!
Hi VK. Thanks for the note! Mi casa es su casa...
DeleteYup. I'm on the same page with regards to multi-channel. There's only so far we can go with 2.0 and that illusion of reality. Of course a great stereo setup can sound fantastic, but if we want to reproduce an even better "illusion", it can be done with more high quality transducers and software.
Heck, even 3.0 with a good center channel can do wonders for vocals - like the Analogue Productions 3-channel stereo Nat King Cole SACDs.
At the real high end I understand they're going back to mono...
DeleteMono. Vinyl. Tubes...
DeleteProgress!
Hi Archimago,
ReplyDeleteCould you test my PCM/DFF/ISO/DSF audiophile converter AuI ConverteR 48x44?
http://samplerateconverter.com
My contacts http://samplerateconverter.com/#feedback
Best regards,
Yuri Korzunov,
Audiophile Inventory
Thanks for the link Yuri.
DeleteHad a quick look at this just now... I see what you're doing with the decoding algorithm.
Will include this with the discussion next time...
Thank you for feedback, Archimago.
DeleteNo only decoding, encoding too.
I like article http://archimago.blogspot.ru/2014/04/analysis-comparison-of-dsd-encoders.html
due there right compared different combinations of encoders/decoders.
I.e. good DSD decoder can't achieve its limits with bad DSD encoder, and otherwise.
Best regards,
Yuri
Time Response
ReplyDeleteHi Archimago
As always, a nice work. Thank you. But what I was missing was some comparison of the time behavior of the DSD to PCM converters. I have also done some comparisons between different DSD to PCM converters and can confirm, that DSD Master and JRiver do a good job with that.
As for comparing the time behavior, I do not use only a single impulse that is on the tentative Test SACD form Sony / Philips, but I use a single cycle 20 kHz burst. With that single cycle burst, I am not confused with different impulse lengths when comparing different sample rate. This 20 kHz has always the same length, no matter what sample rate.
Attached you will find the comparison between the DSD Master Output and JRiver in Safe and one in Medium setting (I would have chosen also JRiver in Medium setting in your frequency file analysis). I have done this comparison not with 96k sample rate (as you have done), but in 176k4 sample rate. So you can't directly compare my results with yours, but you can have an idea.
www.juergenreis.de/DSDMasterDSDtoPCM176.jpg
www.juergenreis.de/JRiverSafeDSDtoPCM176.jpg
www.juergenreis.de/JRiverMediumDSDtoPCM176.jpg
And as for listening test. If you have a good minimum miked recording with a real sound stage, then you will notice differences in the presentation of the soundstage with different time behavior of the DSD to PCM converters.
Juergen
As always, I appreciate the input Juergen. Interesting graphs of the impulse as well.
DeleteCurious Juergen, do you have any suggestions for a good minimum miked recording you've felt can be a good "benchmark" to use as a basis for listening tests?
Even not a pure 2 channel miked recording, but a good one, for comparing different sample rates is for example "quiet winter night" from "hoff ensemble" recorded in native DXD bei 2L, recorded only with 5 microphones, in similar INA5 configuration (no mix between different close mics). For example with this file (a free one is also available on the 2L "test bench" page) you can clearly hear the differences of different DSD64 to PCM96 conversion. PS: I am not a fan of DSD, but I am well aware of the importance of "good" anti-alias PCM filters during recording and during playback and against the standard school book linear phase filter with time symmetrical pre and post ringing. (PS: Those DXD files are recorded with DAD converters at 352 kHz PCM without anti-alias filter, because there is no audio above 176kHz that need to be suppressed during the recording. So the only filtering is during playback and this is the reason, why you will / could be able to hear the sonic differences of different anti-alias filtering during playback). You can also compare your downward conversion results with that of 2L (using saracon (as far as I know)).
DeleteGood luck.
Juergen
Thanks Juergen, when have a look & listen!
DeleteThe question I have is why DSD? DSD is a niche within a niche. There are no sonic benefits, almost all DSD files come from PCM or analog sources. The high frequency noise is lame and above all, PCM has been perfected in the lat ten years or so. SONY is looking for people to buy (again) their music collection in DSD which was designed for archiving. Time will tell but I don't see DSD being too relevant outside of a very limited audiophile market.
ReplyDeleteRan
No argument with you there Ran. I think it's all so obvious to anyone who spends a bit of time thinking about this, taking some time to A/B DSD64 vs. a 24/96, and considering the production side of music.
DeleteThe only benefits are the odd "pure" DSD direct recordings out there usually from small audiophile labels where everything can be kept in DSD form. But even then, only if the playback chain has good low-pass filtering. Even then, who's to say that the quality would not have been better if it was directly recorded to 24/192+ and resampled down to 24/96 or so...
As far as I know at this time, the only people benefitting from DSD are the purveyors of such files in terms of financial markup.
Ran,
ReplyDeleteCurrent wide applying of DSD is not quality matter.
DSD decoder significantly simpler and cheaper native multibit PCM (voltage matrix) DAC.
DSD decoder has 2 voltage levels only and enought slope analog filter.
So it more easy way achieve more expensive multibit DAC quality.
As I hears periodically from different sources, now almost absent consumer's native (voltage matrix) PCM DACs, most "PCM" DACs has DSD modulator-demodulator inside.
Best regards,
Yuri
Indeed. Many DACs are Delta Sigma. The issue I have is not with the DAC but with the file format.
DeleteUsually computer have more significant computer power than hardware devices.
DeleteEspecially almost not limited offline conversion.
So we have "ready for consuming (playback)" stuff.
Hi Archimago,
ReplyDeleteThanks for the artical.
I just spent some time on this DSD2PCM conversion few days ago, and I wanted to find some good settings.
My Foobar2000 settings are:
SACD Plugin:
PCM Volumne: +6dB
PCM Samplerate: 352800
DSD2PCM Mode: Installable FIR (Double Precision) -> Default 30KHz filter
DSP:
Resampler (SoX): 96KHz, Best Quality
Such setting will give me:
Noise Level: 149.4, 147.4
Dynamic range: 133.2, 133,1
Stereo Crosstalk: 148.5, 147.3
I hope this information could be useful for you.
Best Regards,
C.J.
The results are Reference PCM vs DSD2PCM
DeleteHi Japhson,
DeleteHere my researches DSD vs. PCM
http://samplerateconverter.com/content/how-impact-audio-quality-pcm-dsf-conversion-1-bit-dsf-vs-pcm
Best regards,
Yuri
Thanks Yuri and Japhson.
DeleteThe RMAA analysis gets interesting with different resampling and filtering algorithms! I can get anywhere from -135dB noise to -140's depending on what is done...
Thanks Yuri for the link. I see you're using quite a steep filter by default with you converter software.
Hi Archimago,
DeleteYes. My software use manually tuned filters for all combinations input/output sample rates.
All these filters are steep. Both linear and minimal phase.
After each filters improving (now many filters replaced from 2 to 3 generation) I try get feeback from AuI ConverteR's users for undersatnding how linked features changings and ear perception. It is not double blind test, of course, however give some info:
Better hearing result give more suppression of artefacts (from -150 to -160 dB). I suppose it more important for weak fragments of music (levels -50 ... -80 dB).
Therefore need use steep filters.
Of course, most hard in realization such filter for 44 kHz :)
Best regards,
Yuri
Where is the Time behavior? Why mostly only compare the dynamic range, in band noise, etc ... What is the basic point within DSD? Yes, the time behavior. So why ignore the biggest difference between DSD and PCM (timing) in most of those comparisons? I do not mean the ultra sonic behavior, I mean the in band behavior.
ReplyDeleteThe time behaviour in-band (assuming 0-20kHz) is exactly the same as PCM. any waveforms with a steeper rise than a 20 kHz signal contain ultrasonic energy... for which there exists no evidence for their audibility. If you refer to the smallest possible delay, again DSD is no better. The time resolution of PCM is the inverse of sampling rate, multiplied by number of discrete levels. In other words,
DeleteT = 1/(fs*2^n)
Where n is the number of bits.
So for redbook CD audio: 1/(44,100 x 65536) = roughly 0.3 nanoseconds. In practice, due to dithering and jitter, it's closer to a few dozen nanoseconds.
If this is still unclear, have a skim read of this white paper. Think of it this way: if a wave is shifted a small fraction of an arbitrary time unit, the amplitude at each sampling point will change.
Here is a practical test designed to demonstrate this aspect of digital signals: http://forums.stevehoffman.tv/threads/time-resolution-of-red-book-45ns.85436/
45 nanoseconds goes way, way, beyond what is simply adequate. Sampling rate makes zero practical difference in timing resolution. Hope this helps!
Sorry, forgot the white paper link: http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf
DeleteHi Anonymous (even I would prefer a real name to answer to).
DeleteI know exactly what you mean and I am also well aware of your mentioned Lavry paper, as also many other papers and AES articles and papers, that I have seen within the last 20 years. So I can also find other AES papers, for example "The audibility of typical digital audio filters in a high-fidelity playback system" from last year, that goes more into my thinking. All I wanted to say is, that when comparing DSD to PCM converters, that there is more to be looked at, that just the THD, Dynamic Range, Noise, etc, and that there are sonic differences, that do correlate with the timing behavior of the digital filters. No more, no less.
Best Regards
Juergen
You might want to take a look at the refutations of that paper in the comment section on the AES site. Meridian employed practices contrary to their own recommendations for transparency- RDPF dither (which modulates with input) and an unusually narrow filter (500 Hz passband) not found in modern hardware. Additionally, the result only barely met the minimum requirement for statistical significance, in an exceptionally well soundproofed room... these all point toward the audibility of redbook's flaws being vanishingly small, if existent at all.
Delete"that there is more to be looked at, that just the THD, Dynamic Range, Noise, etc, and that there are sonic differences, that do correlate with the timing behavior of the digital filters. No more, no less."
Without any provision evidence, and given the existence of plentiful evidence to the contrary of your agenda, it's hard to take you seriously.
No worries Juergen, I'll have a look at these impulse responses for this coming week for completeness... Thanks for your images above and I can certainly confirm findings like the use of minimum phase filtering with DSD Master as your graph showed...
DeleteSure, multi-MHz sampling would give us very nice impulse response graphs, however, as 'anonymous' suggests, I'm really struggling to come to terms with the notion of actual audibility with music playback...
In the real world though, that multi-MHz sampled signal will be lowpassed at the output of the SACD player or somewhere along the signal chain, destroying that 'perfect' looking impulse response. Not that any of this matters if noone can hear ultrasonics.
DeleteJuergen, I'm posting as anonymous via Google accounts because that's the easiest option. I don't have a blogging account.
Addition to my impulse measurements, that I have shown above. These are measured in digital domain (and not after a possible oversampling filter of a DA converter), so it contains only signal content below halve the sample rate. So when I would have done this with end files of 44k1, then there would only be signal content below 22k05.
ReplyDelete@Archimago: Sometime I do see, that you are showing the time behavior with Adobe Audition. I would like to inform you that these program does not show you the pure samples, it has also a steep linear phase low pass filter. So when you look at a real dirac impulse with Audition, you will see (optically a symmetrical pre and post ringing, that doesn't exist in real). If you are looking more closely, you will recognize the dots, representing the real data points.
www.juergenreis.de/DiracWithAudition.jpg
Juergen
Yes. I'm aware that Audition will render with the linear phase filter "assumed".
DeleteBut is this not reasonable from the perspective of real-world DACs? Other than units that specify use of minimum phase, slow roll-off, etc. or are actual NOS designs, would this not be a reasonable assumption in the vast majority of DACs?
In any case, when DSD conversion of 24/96 data is "rendered" to 352kHz as I'll show in the post this weekend, I think it's pretty clear what the conversion programs are doing with the impulse signal timewise...
Depends on what you are looking for.
DeleteIf you mainly want to have a look, that the Downsampling is doing, then it is better, to look at "clean" data (and not to be mislead, what will be after a "typical" Linear Phase DAC).
You may know of that website, Infinite Wave, where you can see a good comparison of PCM96 to PCM44k1 SRCs, with "clean" time data. http://src.infinitewave.ca/
But when you want to see, what will be at the output of an typical Linear Phase 4-times Oversampling DAC, then you need to "simulate" this in digital domain and then compare. For example good meters for mastering are doing this that way, in order to see an possible intersample overload that could otherwise occur, in typical digital oversampling filters.
When when you are doing this, then you are mixing two effects together. For a clean analysis, I would prefer to show just what the SRC process is doing.
Looking forward to your next measurements and also looking forward, what your sonic impression will be, with the sound file, that I have told to you.
Juergen
Hi Juergen,
DeleteYou keep mentioning sonic impressions, sonic differences and the like. I wonder, have you noticed that the pre and post ringing of all filters occurs only in the transition band? That is, between 20 and 22.05 kHz in the case of redbook audio. In light of this fact, the burden of proof actually lies upon the claimant who asserts that this ringing or time domain 'smearing' is audible. I suggest you try an ABX test with files filtered by various means. I've done so myself and could not detect any difference between minimum phase and linear filtering done in Izotope. But if even a single person is successful in doing so, my stance will change! So give it a go, you may be the first to change people's minds after all. I'm not at all opposed to the idea that redbook is imperfect- most of my own library is digital and hard drive space is plentiful nowdays. But I've yet to find any evidence...
Hi Anonymous
DeleteI am well aware of the fact, that the pre and post ringing, does only occur with FS/2 as frequency and the level is dependent, on how near the audio frequency comes to that FS/2. This point does also struggle me, when listening to different digital filters.
I do not want to convince anyone and anyone can have his own opinion. I am fine with that. I just want to report a bit about my experience, and everyone is well come to report about there experience. Even when we disagree, at the end, if we enjoy listening to music, this is fine.
All I can say, when I have designed digital oversampling filter (in hardware) very extensively last year, that I was surprised what all was audible (for me) so I take my own conclusion (that may differ from others conclusion). And I want to add that, if you take recordings, that are made without anti alias filters during recording (see apodizing articles from Meridian) and having control of the digital filter at the playback side, that this is then the only digital filter in the complete path (from recording to listening). When someone is trying with software filters (like Izotope (they have really good SRC and digital volume control)), then you do have this software filter always in series with the digital filter of your DAC (and maybe a anti alias filter at the recording). I hope it is clear what I am trying to explain (I am not a native US). When I was trying to listen to different software filters, over a dac with a fixed linear phase filter, than the differences was very difficult to hear. But when doing the modifications at the DAC side and hearing only one filter during playback, then the differences where more easy to hear (with good minimum miked recordings and when having to sonic signature of an alias filter during recording).
Maybe the sentence is a bit chaotic, but I hope it is clear enough what I mean. And again: I do not want to convince anyone. I just want to report my experience.
Juergen
Hi Juergen,
DeletePerhaps that is where our philosophies differ. Given that no current model of psychoacoustic science could explain the perception of very low level (-65dBFS at worst case) ultrasonic ringing/Gibbs phenomenon 'artifacts', I believe it's reasonable to demand a more rigorous examination of what people are hearing (or not?). So while I fully agree with you in that everyone should have one's own opinion and enjoy the music above all, I'm unsure what posting unverifiable, non-repeatable, end user-specific experiences achieves when it comes to such a topic.
I don't intend to devalue your subjective experience nor do I dismiss the simple value in just -listening- to the sound. But when evidence is so heavily stacked to one side, it seems a little redundant for one to offer purely anecdotal rebuttal. Blind test are a different story however as we can control variables there.
Hi Anonymous
DeleteI could add some examples from my hobby as musician or recording engineer (mostly classical music), that there is more in what we feel when we are listening to but this would be a far too long story. Just a short example. I do not hear anything above 17 kHz, for sure, so the Gibbs phenomenon, even at CD resolution, must be inaudible for me, but when we play in some concert halls, where they have ultrasonic birds protection devices or / and also fire alarm devices (that work both above 20 kHz), most of the musicians do not feel as relaxed as they do, when the devices can be switched off. There are also some papers about Brain activities with frequencies above 20 kHz, that I do not refer to.
And secondly. It sounds like you have also a technical back ground. So take a minimum miked DXD recording (without anti alias filter) and run it through a programable hardware filter for sample rate and anti alias filter down to 44k1 direkt into a DAC chip with digital filter section by-passed, so that you have only one digital filter in your signal path and listen to is. Make linear phase and minim phase comparisons, change the steepness and change the stoppband suppression (and shape). You will be hear a difference in the sound stage (but don't mix up a software frontend digital filter with a followed backend hardware digital filter in the DAC).
Good luck
Juergen
Juergen,
Delete2-3 years ago I released minimal phase filter for my audio converter software.
I could made enought linear phase response.
Minimal phase filter eliminate pre-ringing. But all pre-ringing energy shifted behind fron of signal. I.e. for minimal phase filter we have 2 x artefact's power.
From my experience almost nobody (what I got as feedback from audiophiles and pro) prefer minimal phase filter.
May be it prejudice?
Me interesting your opinion about comparing linear and minimum phase filter.
Best regards,
Yuri
" but when we play in some concert halls, where they have ultrasonic birds protection devices or / and also fire alarm devices (that work both above 20 kHz), most of the musicians do not feel as relaxed as they do, when the devices can be switched off."
DeleteThis may well be the case but to deter birds the amplitude of these signals (as high frequencies are very directional) must be VERY high.
The ringing of filters is short lived and VERY low level (many dB's below the amplitude of the stimulus).
When I playback a signal above my audible threshold and play it very loud I do feel an uncomfortable 'pressure'.
There has been research showing the brain does seem to react to ultrasonic contents that has a relation to the music signal. Unrelated high frequency content doesn't seem to create the same brain activity. Strangely enough it doesn't seem to correlate with the perception of sound with the same testgroup.
Also research has shown that we can percieve harmonic contents with test tones where harmonics that are above the audible treshold still influence the perception of the tone.
Here too, this has been shown with music, just single high frequency squarewaves with high amounts of harmonics (amplitude wise).
Then again there also exists papers about smoking, cell phones, power lines close by, cell phone towers, global warming and one can find papers that 'prove' there is no influence and papers that state there is a substantial influence.
Trust only your own ears not those of others... if 'it' does something for you then you can enjoy the music more, if it is all moot, a storm in a glass of water and you can't hear any of it then don't bother with all of this and simply enjoy the music, regardless of filtering.
Obviously filters (44.1/48) that show a considerable roll-off in the audible range can easily be picked in blind tests.
Which of those filters person A or person B prefers may be more of a personal matter (due to the roll-off) then determined by ringing artifacts.
I find it interesting, where this discussion is going to, and I find it very good, that this discussion is without personal attack and is going a positive and constructive way. When it comes to what is audible, and what is not audible, on some forums they are "shooting" against each other (and this would be waste of energy and time).
DeleteYure, what you have written about preferences to linear phase correlates with my impression, when I first heard a digital minimum phase filter, after I was used to listen to linear phase filters the other 10 years before that. But at least the group of persons you have asked, have heard a difference in the sound (even this ringing is above our hearing abilities).
When I would mix and master with listening to / with a Linear Phase Filter DAC, I would do it different to when I am mixing the same recording with listening to / with Minimum Phase Filter DACs. So if it is mixed to sound good with LP Filters, it would at least sound different with MP Filters (and I am using mostly minimum miked technique).
Let us see, what Archimago will measure and public, in his next round of DSD to PCM conversion.
Juergen
Juergen, thank you for reply.
DeleteI suppose, user of the filter must have choise ability (linear/min.phase). And select what he/her more like.
Best regards,
Yuri
This comment has been removed by the author.
DeleteSome good points being made, but we must be careful not to equate 'trust your own ears' to 'ignore all variables because my own hearing reigns supreme over cognitive biases'... If we intend to draw any valid conclusions, that is. Otherwise, yes, just enjoy the music. :)
DeleteSolderdude Frans, would you be able to link those studies you mention? Or just link the abstracts. Thanks in advance.
Deletehttp://en.wikipedia.org/wiki/Hypersonic_effect
Deletehttp://www.nhk.or.jp/strl/publica/labnote/lab486.html
http://jn.physiology.org/content/83/6/3548.full
http://scitation.aip.org/content/asa/journal/jasa/122/3/10.1121/1.2761883
The 'problem' with many of these studies is that conclusions are often drawn on the average amount of people, often this will be around 50% 'proving' statistical BS.
In some studies critical listeners are used in others non critical listeners.
It may well be possible that those that can accurately detect hypersonics and have proven it in tests are 'dismissed' as statistically irrelevant with a chance that someone guessed correctly.
In the end ... if someone is convinced they can hear it they should do everything they can to get maximum sound quality.
If one cannot hear it then don't bother. I am one of those as I already get enough enjoyment of well recorded 16/44.1.
Personally I think people want to hear differences more so than they actually can.
KNOWING what filter (or equipment) is in use invalidates listening tests IMO.
Hard to test this blind yourself without the help of others.
To me the obvious audible differences are seen with slow roll-off filters and NOS DAC's that lack proper reconstruction filters and have roll-off in the audible range.
It's that 'talk' of preferred sound that (to me) is responsible for most of the 'talk' out there.
With everything in life ... one should go with what one feels is correct.
I think it's a great thing people want to improve audio and that it continues to be an ongoing effort.
> but we must be careful not to equate 'trust your own ears'
DeleteI am, as example don't trust my ears. It's very unstable tool.
Me need work many hours everyday. Ears are tires.
Measurements is simplest way check moving to right direction in development.
Ah yes, I fully agree with that statement.
DeleteYou can NOT trust the ears/brain as a 'measurement tool' at all.
They should only be used for enjoying music (in relation to audio).
I am ALL for measuring as much parameters as needed, to satisfy my technical curiosity or to see where something can even be improved technically.
I believe that audible differences are always measurable but not all measurable differences are audible.
When there are audible differences and they don't appear in certain measurements than one either has not measured the parameters that matter or the the brain may have been fooled.
I have made people hear 'obvious differences' by simply telling them I changed (this or that) while in reality changing nothing. This also tells me NOT to trust your hearing as a 'tool' as it can easily be influenced by lots of other 'factors/circumstances'.
It is all about audibility thresholds and knowing one's limitations therein.
In the end one ends up using their own ears when enjoying/listening to music, and that's what matters, it should not be the opinion of others or test results/numbers.
So in that context (of enjoying music) I say 'trust your own ears' and not those of others nor their opinions.
When I played in band many years ago, not once was same thing:
DeleteIn evening we (band) got (after tuning of apparatus) perfect sound and got many many drive.
Next day morning sound was poor without any changing of settings of apparatus.
I suppose nice mood helped perceive sound. Also health, room, speakers placement. Here many factors.
Sometime I hears ununderstanded for me things, that unexplained from technical point of view.
As example, no one men said about better sounding WAV than FLAC.
I personally check full binary identity WAV=FLAC http://www.youtube.com/watch?v=3jPphh-CsHM
We can consider third party players as black boxes.
Here can assume some DSP for FLAC only, or light level difference or used FLAC library error.
But both things low probable.
Same matter: why some people hear differency between CDs from different manufacturers.
I don't deny these matters (currently unexplained from technical point of view). Possibly here reasons that we don't know (technical, of course). Possibly not.
I wouldn't like to have to feed the amount of people that claim to hear differences between FLAC and WAV.
DeleteThis has been covered on this blog already:
http://archimago.blogspot.nl/2013/05/measurements-do-lossless-compressed.html
Those that make the claims either have faulty or poor performing equipment or have some kind of poor conversion stage/player/OS setting.
The same goes for differences between CD's.
How can we be sure that the CD's in question have the same master or that a player has problems reading the 'copy' ?
There is no reason to believe that there are aspects that are yet unknown.
For everything there is an explanation, it's just that not everyone likes to hear it.
Yes... this goes for both sides of the viewpoint.
Of course this isn't really on topic any more and this discussion belongs in forums and such.
This topic is about DSD to PCM conversion and the audibility of this given the (BS) picture above and the supposed lack of speed PCM has opposite DSD.
For checking reasons of possible difference need deep learn details. May be man who say about difference don't said some important details about apparatus/player/conversion software settings.
DeleteAbout CD I also listened that same CDs from different manufacturers sound differently, but ripped both sounds identical. I tried discuss this matter for detecting possible reason. But while don't get suitable ideas for explaining.
In audio I'm dry materialist :) However keep open minds for same things. May be sometime it will explained. May be difference's reason in some thing that we don't suspect before. May be no difference.
You would have to analyse the CD output (read-head level) and check for the amount of C1, C2 and uncorrectable errors. (very expensive device)
DeleteCheck for jitter as well.
IF there is really a difference this test equipment can tell.
Or you can do the test Archimago did (recording both CD's and null them) which is the cheapest way AND is done at audio levels.
Or you could repeat the listening test truly 'blind' and or use another CD player.
I suppose uncorrectable errors (by theory of probability) give clicks/pauses only.
DeleteClicks - when error is not detected - player consider it is right data.
Clicks - when error is detected and wromg corrected - player consider it is right data.
Pauses - when error is detected and can't corrected.
Loss transparency (as I understand many people said about it) can has 2 reasons:
1. Jitter by reading via optical system [CD-laser-optical receiver]. But it eliminated via internal FIFO buffer that has clock by CD player. Here jitter defined by CD player, not CD-disk.
2. Periodical (period 0.1 ... 1.0 second) pauses or undetected errors.
I 2-3 times stumble with case 2. However, if disk so damaged, sound is worse than simple loss transparency.
Errors in data are as good as always detected, they rarely go un-noticed as various types of safeguards (EFM, interleaving) and error correction techniques (reed-Solomon, parity) are used.
DeleteC1 errors are constantly present and all CD's have them, some much more than others, these are correctable.
Most people think 'corrections' are always in hindsight and thus not perfect but this is not the case at all. The incoming data is checked and corrected by calculation based on the CIRC technigue and only then fed forward to the digital filtering and lastly the DAC chip.
Lots of calculations/time/clockpulses between the time the data is read, decoded, processed and sent to the DAC. CD playback is NOT real time like analog audio.
C2 errors are occuring less often and also correctable in most cases.
When not correctable by calculation these errors are flagged that way and usually the missing sample is held at the last sample value or interpolated with the previous and next verified sample value(s).
This isn't really error correcting any more though, but error concealing.
When this happens constantly/often the sound quality degrades before it becomes obvious as ticks.
Uncorrectable errors CU are handled by the firmware in the player and is chip/manufacturer dependent. Interpolation, sample hold, glitches, muting or holding last position is possible and a designers choice.
In that case either the player needs adjustment, is defective or the disc is damaged or incompatible.
Not all CDP's play CD-R that well, most don't even play CD-RW... some do.
Depends on many aspects.
As you already mentioned the read speed of the disc is determined by the master clock in the CDP.
The DAC chip on board is not clocked by a PLL, like external DAC's, but are governed by the master clock.
If the clock in a CDP has loads of jitter than it is also in the audio regardless of disc quality.
There are different kinds of jitter though and not all are audible even in substantial amounts.
Jitter can be measured (at the eyepattern level) with very expensive test equipment and should be verifiable.
Some claim tracking and focussing 'action' (because of poor PCB design) can influence the clock or timing of digital circuits or even the analog signal. The more physical correction of the lens is needed/present (tracking, focussing) would result in degradation of sound quality.
Error correction in the digital plane is NOT the same as the 'error correction' of the laser beam desperately trying to follow the path of the land/pits (or ATIP for CD-R)
VERY easy to check for the latter by burning a disc with no audio in it and playing that back while messing (tapping on) the player etc.
If only steady noise comes out (background noise of the DAC/post proc) then that cannot have any influence of the sound.
I have never experienced the findings you speak of but know others have reported them as well.
Probably the same people who hear differences between FLAC and WAV, differences in ethernet cables etc.
Perhaps one shouldn't assume degradation of sound quality can only have 2 possible causes.
It might have other technical causes or perhaps may not caused by any technical problems at all but are of another nature.
I wish everybody happy hunting for these 'anomalies' and hope someone may actually find some technical answers to these questions that keep the minds of some audiophiles so occupied.
Personally am not going to look for it (any more) unless there is some really compelling technical evidence that points in a certain direction. I would rather spend my time enjoying music.
I would like to leave these of topic ramblings for what they are, as for now it is just a guessing game.
The same dirac impulse, now shown within Reaper:
ReplyDeletewww.juergenreis.de/DiracWithReaper.jpg
Juergen
Even I do like Audacity not as much as Audition, it does show the dirac impulse "correct" as it does Reaper.
ReplyDeletewww.juergenreis.de/DiracWithAudacity.jpg
Juergen
Juergen,
DeleteYeah, I'll have a listen to "Quiet Winter Night" as suggested above. Likely will not have time to fully explore the subjective characteristic impression this week but will at least get the measurements and graphs done for these DSD-to-PCM converters to get that out.
I've seen that InfiniteWave page over the years as well. I didn't know until looking over the FAQ that Alexey Lukin was involved in those measurements as well as development for the RightMark Audio Analyzer and iZotope RX which I use regularly (and which I'll be using with the next blog post).
iDealshare VideoGo can convert between various video and audio formats. It support batch conversion. fast speed, play, edit funtion.
ReplyDeleteWindows version http://www.idealshare.net/video-converter.html
Mac version http://www.idealshare.net/video-converter-mac.html
If you want to free download the alternative to Foobar2000 for Mac andindows to play MP3, MP4, AAC, CD Audio, WMA, Vorbis, Opus, FLAC, etc and convert between any audio formats.I suggest Avdshare Video Converter,You can find it at http://www.avdshare.com/foobar2000-for-mac-and-windows-alternative
ReplyDeleteIf you want to convert all music, DSD, FLAC, mp3, aac and so on with the best audio quality I suggest you try the EZ CD Music Converter. It has free download available.
ReplyDelete