Anyone else upgrade and stop hearing 24-bit advantage?

Last night I changed the algorithm that resamples 16/44.1 to 24/96 on my linux computer before sending it to the Wavelength Proton USB DAC. I changed from "samplerate_best" to "samplerate_linear" and I can't believe the improvement. I think it's a simpler algorithm and definitely much less CPU-intensive.

I was planning on starting a major needledrop habit in order to feed my 24-bit fix, but now I'm getting the same quality of sound from 16/44.1 and native 24/96. Has this happened to anyone else? I can't believe how much can be squeezed out of redbook. There is a huge improvement in both detail and musicality.

 

If you're hearing a difference with samplerate_linear than it is purely due to added distortion (specifically aliasing and a treble rolloff). If you like the distortion, and it makes 16/44 sound as good as 24/96, then I guess you've answered your own question.

Such a situation hasn't happened with me.

 

Wow, that's amazing. Can anyone corroborate this? I didn't want to say anything before to avoid sounding cliche, but it sounds incredibly analog.

It's a large warmness/smoothness/musicality boost as well as detail. I'm telling you I'm hearing lots of stuff I've never heard before in *every passage* of *all* of my music, plus a big increase in emotional listenability. Imaging, etc. I'm loving this.

EDIT: Publius, have you ever heard samplerate_linear?

 

I don't need to. Math doesn't lie. That the "linear" conversion setting will have vastly higher distortion than the "best" setting is a foregone conclusion, because linear interpolation inherently has worse passband and stopband performance compared to a sinc-based interpolator, resulting in both attenuated high frequencies and aliasing.

What you hear is your own business, and if you like the sound, that's fine. But it is simply far more probable that such a difference in sound, if existant, is due to such added distortion. It might be due to other factors, of course, but this is far and away higher in magnitude than anything else that is likely to happen.

It is worth noting that the documentation for the resampler being used describes the linear converter as: "the quality is poor, but the conversion speed is blindingly fast."

EDIT: And the converter is featured here as "Secret Rabbit Code 0.1.2 (Linear)".

 

It may very well be distortion (it also may not) but vinyl and tubes are said to add distortion. Do you like the sound of those?

 

I listen to vinyl, if that's what you're asking.

Sorry, I know that my OP was a bit of a threadcrap, but I found it very important to clarify that many people would characterize moving from a sinc interpolation to linear interpolation as anything but an improvement - as others may not be aware of the very real controversies involved here. Essentially, you're moving along a notional axis of DAC design which terminates, in a sense, with NOS.

I'd be curious as to what your thoughts are about samplerate_order.

 

I'm sure you see my point.

Could you clarify this a bit? I'm familiar with the concept of the non-oversampling DAC.

Sounds bad. I tried samplerate_best, samplerate_medium, samplerate, samplerate_linear, and samplerate_order. I'd been using samplerate_best for a while, then I tried samplerate and found that it sounded better. Moving in the same direction, I tried samplerate_order and samplerate_linear and I was blown away when I heard the latter. I still am.

This isn't a sound preference thing, it just sounds much much better. It honestly sounds analog. It could certainly have to do with my equipment, but definitely not my ear.

 

Hi ggking7,

I have yet to hear a real-time sample rate conversion algorithm that does not add spurious harmonics, which can easily be mistaken for "more detail". In fact most (but not all) of the off-line SRC algorithms I've heard do the same thing.

To answer your question, no, I continue to hear distinct advantages, across the board, when recordings are digitized at 24-bits and high sample rates instead of 16-bits and low sample rates, even with the best 16-bit playback.

Just my perspective.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Barry, have you tried resampling 16/44.1 to 24/96 with samplerate_linear before feeding a 24/96 DAC?

I'm hearing so much new stuff in all my music. I can understand lyrics I couldn't before, I hear someone shout in the crowd that I couldn't before, and each notes rings with intricacies when before it just stopped. It isn't a hard-edged detail-fest though. Somehow the smoothness and detail have been exposed simultaneously.

There may still be a difference between 16/44.1 and 24/96 to my ears, but the gap has been drastically narrowed. That wasn't really my point even though I put it in the title. What's exciting is how much better Redbook sounds. I didn't think there was this much info in there.

EDIT: Could samplerate_linear sound better than the more CPU-intensive algorithms because it's faster and thereby less prone to jitter?

 

For the rest of the class that came in late, I take it this is a discussion on the merits of resampling 44/16 to 96/24 (or is it 96/16)?

The settings "samplerate_linear" and such are linux ALSA driver settings?

And the proposition is that "samplerate_linear" sounds nearly as good as native 96/24?

Forgive my respectful cynicism.

 

I should have written a better title and OP for this thread, but it's a discussion on the merits of resampling 16/44.1 to 24/96 via the samplerate_linear algorithm *on my system*. I don't believe that this is any sort of a "sound preference" or "sound naivete" issue of mine at all. I do believe it could have to do with my system though. Maybe the Wavelength Proton's true glory is finally shining upon me, now that I've simplified the ride music takes before it arrives with her. I have to say though, moving to the Proton with the old configuration from the Monica USB DAC was a very pleasant upgrade also. This takes it to the analog level though. Thoroughly pleasing sound. I haven't mentioned it yet, but also much deeper and defined bass. The timing or something has changed also. The whole thing rolls and bounces along like I haven't heard outside of vinyl.

Yes, but they refer to the "Secret Rabbit Code" family of algorithms that are also available on Windows.

Yes.

Yeah, it's the kind of crap I used to read, snicker, and move on from. This has been going on for 2 days and I can't believe I'm still being subjected to this wonderful sound.

 

Hi ggking7,

So far, I've tried several dozen SRC algorithms but none have been for Linux. (I've tried hardware SRC and Mac software SRC as I use a Mac for my audio work. I've yet to hear competition for iZotope's 64-bit SRC. But I have not yet heard the one you describe, nor have I seen any tests of it.)

I can't say for sure but some of this may be a function of gentler filtering in your DAC at the higher sample rate.

If it sounds better (in this case meaning more like the unconverted original as opposed to what one likes), I would imagine it has more to do with better math and less generation of spurious harmonics.
But since I don't know the algorithm, I can't say for sure.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Hi ggking7,

If this is indeed related to Secret Rabbit, I think iZotope's 64-bit SRC will knock your socks off. For my ears, there is no contest between them. (Again, my criterion is not my "liking" the result but the creation of a result that sounds the most like the unconverted original.)

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

One of the things I love about this forum is learning new things, or learning about things I think I should have already known about. I just heard about Secret Rabbit Code for the first time today.

 

ggking, I would highly recommend using sox, and this should be easy for you since you use Linux and sox can resample on the fly.

I do needledrops 24/96 and resample as such;

cd to the directory with your 24/96 wav files

Code:

mkdir -p 441
for i in *.wav; do
 	sox $i -b 16 441/$i rate -v 44100 dither -s 
done

If you have a few brickwall peaks, sox might barf on some of your samples (it will throw an warning messages and leave a glitch in the resampled file). In that case you might want to tweak down the gain to give a little headroom:

sox $i -b 16 441/$i gain -0.2 rate -v 44100 dither -s

Gain is in dB, so -0.2 dB is very small - probably inaudible.

sox at very high quality does an excellent job. I can't find that site that compares all the different resampling software with response curves, but sox with rate -v is clean as you can get.

I have no explanation as to why a poorer resampling algorithm would give you a better sound. I suspect something else is going on.

Oh, one more thing: I used to use ssrc (which uses the 'secret rabbit code') - its characteristics are not as good as sox with very high quality, so I stopped using it.

Ah: Here is the site I was thinking of which compares all the resampling software out there: http://src.infinitewave.ca/

 

Well, it is secret after all!

 

iZotope's 64 bit steep/no alias filter looks identical to sox vhq linear phase except iZotope rolls off a little more gently as you approach 22.05 kHz

 

ggking, it occurs to me that perhaps you did improve the sampling rate algorithm without knowing it... could linear refer to the phase, which for linear means you don't have frequency dependent phase errors upon resampling? What library/software are you using to resample? If it's alsa, samplerate_linear is definitely not "supposed" to sound good as samplerate_best!

 

I believe it's this

 

Hi Leigh,

I find the sweep and even more so, the 1k tests to correspond quite well with what I hear in different algorithms. The 1k test gives an idea of how much brightening and hardening will result from an algorithm. Look for the harmonics in the iZotope steep, no alias setting.

Sox is good, especially for the price. Probably better than most.
But for me, iZotope's SRC is in a class by itself.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Well, I was upsampling to 24/96 before, it's just that I was using samplerate_best.

Leigh, I upsampled from 16/44.1 to 24/96 via sox with your instructions and it does sound very good. I'm going to do some testing. Does the sox command you've typed include the best settings?

Does anyone know if there is a way to use iZotope in linux?

I found this:

http://www.mega-nerd.com/SRC/faq.html#Q005

and I'm going to contact the ALSA guys about it. I'm resampling in ALSA via dmix, but I've also compared the algorithms when the resampling takes place within mpd with the same results.

 

-v is very high quality setting, so it's the best setting. The dithering in my command line employs noise shaping (-s flag). Wikipedia has a good entry on the noise shaping technique if you are interested.

By default sox uses linear phase, which has equal pre- and post- echo, which seems to be the best compromise.

You can employ the secret rabbit code algorithm or the polyphase algorithm with sox if you want to experiment. I don't worry too much about this since my vinyl rig is hardly world class and I figured distortion etc. with my setup will overwhelm such subtle resampling parameters.

 

Leigh, weren't you saying the sox resampler is better than SRC? It sounds like you're saying it would be an improvement to have sox use the SRC algorithm.

Is dithering necesary when resampling up: 16/44.1->24/96?

EDIT: Do you know if ALSA's dmix can be configured to resample with sox on the fly? I can make it use SRC. Maybe all I need is a rate_converter name for high quality sox?

 

No. In this application, you're adding bits, not taking them away. In other words, you're expressing 16 bit quantities in 24 bits. You only need dithering if you go the other way.

 

To (try to) answer your first question, my switching over to sox was completely due to the sweep response I saw at the link I posted earlier. (http://src.infinitewave.ca). ssrc seemed to put out more harmonics than sox hq. This isn't based upon my ears. ssrc sounded fine when I used it. But again, I'm recording vinyl from a mid-level rig. I only mentioned the option of trying the src and polyphase algorithms because I was reading the sox man pages that said these options were available, and I though you might want to give them a shot.

I do not know how to configure alsa to use sox as a filter, or whether it's possible. I don't understand your setup. I'm a squeezebox guy who is downsampling to 16/44.1 because that's all the squeezebox can handle... the next generation version coming out in a few months will be able to pass 24/96 and that may change my strategy... or not, as I'm not convinced it's worth it for needledrops except for special circumstances (beyond the realm of this discussion). So I am simply streaming flac files that have already been converted. It sounds like you're doing something different.