Hi everyone,
I am in the process of purchasing a 24 bit 96kHz NOS DAC (Metrum acoustics). In many ways this is one of the few true 24 bit NOS dac. Many other DACs introduce upsampling within their chip/chipset. Metrum uses an industrial rather than an audio DAC.
I am looking around if there is any software that could oversample a 16 bit 44.1kHz WaV file to 24 bit 88.2kHz FLAC.
I would like to emphasize, I don't quite like the upsampled sound. Over sampling in this context means taking the (V1+V2)/2. Upsampling is curve fitting, and sometimes result in a voltage that is not in between V1 and V2
(which is why i generally don't like it) and in many ways reconstructing the waveform to fit an nice sine wave.
Many high end DACs have a selectable variable reconstruction filters, which tends to lend itself to a different sound, and the user can select the reconstruction filter desired. I am also open to a software that can employ a variable reconstruction filter (sometimes just called a selectable digital filter) on 16 bit 44.1kHz and save them in a higher resolution file (24 bit, 96kHz or more).
I was considering J-River, but not sure if it has this kind of function.
Any feedback will be greatly appreciated.
Oon
I am in the process of purchasing a 24 bit 96kHz NOS DAC (Metrum acoustics). In many ways this is one of the few true 24 bit NOS dac. Many other DACs introduce upsampling within their chip/chipset. Metrum uses an industrial rather than an audio DAC.
I am looking around if there is any software that could oversample a 16 bit 44.1kHz WaV file to 24 bit 88.2kHz FLAC.
I would like to emphasize, I don't quite like the upsampled sound. Over sampling in this context means taking the (V1+V2)/2. Upsampling is curve fitting, and sometimes result in a voltage that is not in between V1 and V2
(which is why i generally don't like it) and in many ways reconstructing the waveform to fit an nice sine wave.
Many high end DACs have a selectable variable reconstruction filters, which tends to lend itself to a different sound, and the user can select the reconstruction filter desired. I am also open to a software that can employ a variable reconstruction filter (sometimes just called a selectable digital filter) on 16 bit 44.1kHz and save them in a higher resolution file (24 bit, 96kHz or more).
I was considering J-River, but not sure if it has this kind of function.
Any feedback will be greatly appreciated.
Oon
Last edited:
Korg AudioGate should do the trick but IME, it is best to play a track in its native format rather than convert the file to something else. I would think that your Metrum dac would be able to play 16bit 44.1kHz files without any issues.
You can download AudioGate for free but you will need to have (or create one if you don't) a Twitter account to use it. If you don't want to open a Twitter account, then you will need a licensed copy of AudioGate. To get this, you must purchase a Korg recording unit.
You can download AudioGate for free but you will need to have (or create one if you don't) a Twitter account to use it. If you don't want to open a Twitter account, then you will need a licensed copy of AudioGate. To get this, you must purchase a Korg recording unit.
I found an interesting one from secret rabbit code that might do the trick, but I have no way of verifying this.
Secret Rabbit Code (aka libsamplerate)
I also came across this really interesting website, that shows the effect of the various kind of filter.
SRC Comparisons
Playing with the impulse response of the secret rabbit code 0.1.2 (linear) results in a triangular wave, suggesting that it is probably doing that.
Oon
Secret Rabbit Code (aka libsamplerate)
I also came across this really interesting website, that shows the effect of the various kind of filter.
SRC Comparisons
Playing with the impulse response of the secret rabbit code 0.1.2 (linear) results in a triangular wave, suggesting that it is probably doing that.
Oon
I don't know for sure. Only going by the manufacturers specification.
I think I didn't make it clear in my first posting. "few true 24 bit NOS dac" in the sense that there is no additional processing done by the DAC chip itself rather than it is 24 Bit. All 24 bit chips perform some form of DSP on the signal (however I have seen some articles on efforts to defeat it).
Oon
I think I didn't make it clear in my first posting. "few true 24 bit NOS dac" in the sense that there is no additional processing done by the DAC chip itself rather than it is 24 Bit. All 24 bit chips perform some form of DSP on the signal (however I have seen some articles on efforts to defeat it).
Oon
Sure. Lots of software around. One of the best tools is SoX - and it is free.
But just to be clear - you are just replacing upsampling in the DAC with upsampling in the software. Who do you think one is better than the other?
That is a very interesting definition of "oversampling". Usually oversampling means sampling at a rate significantly higher than the Nyquist rate. Your "(V1+V2)/2" is linear interpolation, and is not a very good interpolation method for audio.
Yes, upsampling is interpolation, just like your "(V1+V2)/2" linear interpolation. Again, linear interpolation is a very crude and limited form of interpolation, and there are much better and more advanced forms of interpolation for audio use.
SoX is your tool.
I've stumbled upon iZotope RX 3 on the internet. Performs all kind of resampling (up to 192Khz) and dithering (up to 24bits) on digital audio files, also declipping which is quite convincing. 
I've downloaded the trial version, which installs nice and works fine overhere (quad core AMD and 64b Win7). This demo is fully functional, but it will not save your files.
It is an expensive toy: $349,-
I've downloaded the trial version, which installs nice and works fine overhere (quad core AMD and 64b Win7). This demo is fully functional, but it will not save your files.
It is an expensive toy: $349,-
if you want only resample try this software (not real time , you have to save files) : hd audio remaster
Cirlinca | High-Definition audio software | FLAC, DVD, Blu-ray
Cirlinca | High-Definition audio software | FLAC, DVD, Blu-ray
+1So does SoX, but it is free, so it clearly can't be audiophile quality
add the lack of GUI and you covered all.
Ah, yes. For that there is audacity, but that is also free, so can't be any good either.
Dear Julf,
I think a lot of confusion and disagreement (NOS vs OS) is based on many misunderstanding about what is the desired outcome.
First of all, what NOS is really after is not NOS itself, but rather to bypass the digital filter. oversampling per-se does not really make the sound less desirable. It is the brickwall digital filter that is the problem, However, it is pretty much impossible to bypass the digital filter and just provide oversampling, they are normally intergrated together. And most DACs, with the exception of the TDA154X series from phillips has built in the digital filter in the DAC chip itself. A short interview of the original proponent of the NOS DAC himself.
TNT-Audio inter.views Kusunoki San [English]
Secondly, the upsampling in many people's mind create a good interpolation between two points, which is a good thing, the line looks smoother with less pronounced square edges. The problem is, upsampling is really more a sinewave reconstruction filter than it is a nice interpolation. That is why when produced with an impulse response, the eventual outcome is a short burst sinewave.. That means if point 1 is 1V, point 2 is 2V. You would expect it to produce a series of voltage between 1 to 2 V to fill the gap. However, with most a sine wave reconstruction filter, you could wind up with 0.5V and 2.5V. Not exactly in between, it is just trying to fit a sinewave into these points.
A 15kHz sinewave signal at 44.1kHz sampling will actually give you a 22kHz signal with a 7kHz beat if drawn on paper, simply because there is not enough points. But a nice digital filter will actually give you a nice 15kHz sinewave. So in other words it is reconstructing what it thinks the original waveform is. How accurate is the reconstruction, that is anybody's guess, because all sinewave will be produced beautifully. I've never seen anybody test a DAC with a sawtooth wave or a triangular wave yet.
I think this article can explain this concept much better than me..
http://www.audioholics.com/audio-technologies/exploring-digital-audio-myths-and-reality-part-1
My objective is take away the reconstruction part from the DAC itself and use software to control it. So in essence just upsample the 44.1kHz signal with the digital filter you want, save it as a FLAC and play it back, this will take away the control from the DAC back in your hands. The easiest way that could connect 2 points is just plain old linear interpolation. However mathematically there are better ways, polynomial interpolation etc... However, I am not sure which filter produces what..
Oon
I think a lot of confusion and disagreement (NOS vs OS) is based on many misunderstanding about what is the desired outcome.
First of all, what NOS is really after is not NOS itself, but rather to bypass the digital filter. oversampling per-se does not really make the sound less desirable. It is the brickwall digital filter that is the problem, However, it is pretty much impossible to bypass the digital filter and just provide oversampling, they are normally intergrated together. And most DACs, with the exception of the TDA154X series from phillips has built in the digital filter in the DAC chip itself. A short interview of the original proponent of the NOS DAC himself.
TNT-Audio inter.views Kusunoki San [English]
Secondly, the upsampling in many people's mind create a good interpolation between two points, which is a good thing, the line looks smoother with less pronounced square edges. The problem is, upsampling is really more a sinewave reconstruction filter than it is a nice interpolation. That is why when produced with an impulse response, the eventual outcome is a short burst sinewave.. That means if point 1 is 1V, point 2 is 2V. You would expect it to produce a series of voltage between 1 to 2 V to fill the gap. However, with most a sine wave reconstruction filter, you could wind up with 0.5V and 2.5V. Not exactly in between, it is just trying to fit a sinewave into these points.
A 15kHz sinewave signal at 44.1kHz sampling will actually give you a 22kHz signal with a 7kHz beat if drawn on paper, simply because there is not enough points. But a nice digital filter will actually give you a nice 15kHz sinewave. So in other words it is reconstructing what it thinks the original waveform is. How accurate is the reconstruction, that is anybody's guess, because all sinewave will be produced beautifully. I've never seen anybody test a DAC with a sawtooth wave or a triangular wave yet.
I think this article can explain this concept much better than me..
http://www.audioholics.com/audio-technologies/exploring-digital-audio-myths-and-reality-part-1
My objective is take away the reconstruction part from the DAC itself and use software to control it. So in essence just upsample the 44.1kHz signal with the digital filter you want, save it as a FLAC and play it back, this will take away the control from the DAC back in your hands. The easiest way that could connect 2 points is just plain old linear interpolation. However mathematically there are better ways, polynomial interpolation etc... However, I am not sure which filter produces what..
Oon
Last edited:
Dear Oon,
I am not sure your message really cleared the confusion.
To properly reproduce a square wave you would need infinite bandwidth. Do you think we hear the harmonics that are above 1 MHz?
What matters is not where the upsampling and filtering happens, but what algorithms are used and how well they are implemented. DAC designers have done a fair bit of hard work to implement the algorithms they use. If you think you can do better, I suggest you try the different algorithms provided by SoX, and see if you can hear a difference in a double-blind ABX.
I am not sure your message really cleared the confusion.
The reconstruction filter that is an essential part not just of the upsampling but of the digital-to-analog conversion itself is an interpolation. If done properly, it is definitely "nice".
No, there are enough points. A 15 kHz sine wave at 44.1 kHz sample rate will produce a 15 kHz sine wave, plus higher frequency components (above the Nyquist frequency). It is the job of the reconstruction filter to remove those extra components, leaving the signal itself. If you leave off the filter, you are leaving off the whole interpolation part.
So will a "nice" (well designed) analog filter too.
And there is a reason for that. A square wave is a synthetic signal has harmonics above the Nyquist frequency. Fortunately we don't listen to square waves.
Really interesting article that explains it very well, but then chucks in "conclusions" that are purely speculative and not derived from the rest of the article.
To properly reproduce a square wave you would need infinite bandwidth. Do you think we hear the harmonics that are above 1 MHz?
And what algorithms do you think the DAC designers use - and why?
What matters is not where the upsampling and filtering happens, but what algorithms are used and how well they are implemented. DAC designers have done a fair bit of hard work to implement the algorithms they use. If you think you can do better, I suggest you try the different algorithms provided by SoX, and see if you can hear a difference in a double-blind ABX.
The above described behaviour is absolutely correct. It is important to realize the digital stream correctly reconstructs waveforms that were bandwidth-limited to fs/2. There is a brickwall filter before the AD conversion (actually also using oversampling, but the implementation is not important). The input filter makes sure the key condition of digital sampling (max bandwidth of fs/2) holds.
Therefore you have to ask - how did a fs/2 limited input signal look like which produced digital data of a single non-zero sample (your impulse)? Well, it looked just like your correct DAC output - "short burst sinewave". The same holds for the popular square wave. In proper setup (which is done when recording audio) an analog square wave signal could never reach the ADC and be sampled. The "preringing" and "postringing" signal so disliked by NOS proposers is the original input and correct output for a "square wave" digital signal.
IMO this confusion is caused by dumm audio softwares which display waveforms by linking the digital samples discrete in time with straight lines, instead of proper oversampling the connecting trajectory to the horizontal screen pixel resolution of the sampling period. If they did, you would never see a square wave or a straight impulse on your screen in the first place and you would never expect to see that on your scope.
Therefore the antialiasing filter behind the DA conversion (either analog, or digital, or both) is inseparable part of the digital sampling chain. The NOS technology is technically wrong. It may sound different (better to someone's ears), but it is just wrong. BTW I have never seen any field outside of audiophilia to propose/promote such mutilation of the digital sampling chain.
+1 - just as bad as their other major error, showing the digital data as a staircase.
Hear hear!
Julf, you asked me to show real measurements, in another thread. So here we go, as it belongs rather here.
The test signal is a TPD dithered 1kHz sine in a 44.1/16 format. Attached plots show:
1) the dithered 1kHz played through a real world DAC when this DAC is driven with these 44.1/16 data
2) the 44.1/16 is first SW oversampled to 96/24 and then it goes to the same real world DAC as 96/24 data.
Quite a difference, what do you think. I agree it depends on the DAC used, but there is no simple general answer.
This DAC is clocked at multiples of 48kHz and uses internal resampler for 44.1kHz data.
The test signal is a TPD dithered 1kHz sine in a 44.1/16 format. Attached plots show:
1) the dithered 1kHz played through a real world DAC when this DAC is driven with these 44.1/16 data
2) the 44.1/16 is first SW oversampled to 96/24 and then it goes to the same real world DAC as 96/24 data.
Quite a difference, what do you think. I agree it depends on the DAC used, but there is no simple general answer.
This DAC is clocked at multiples of 48kHz and uses internal resampler for 44.1kHz data.
Attachments
Last edited:
Thanks!
Yes - seems the upsampling in the DAC is somewhat less than perfect
I agree - it is all DAC-dependent. One can not make general claims of the "it is always better to upsample in software". It is always a question of "is my software upsampling better than the one in the DAC?". In some cases yes, in some cases no.
As I keep saying, the best way to find out is to use SoX (or Audacity) to try - the SoX algorithms are pretty darn good, and it doesn't cost anything to try. The most demanding part is arranging for proper double-blind ABX conditions.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.