I got me a fancy new Samsung Galaxy S5 phone. The fine print states that the DAC in the device supports 24-bit/192kHz audio. Google's Play Music app (their default music player) supports FLAC. I copied some 24/192 albums in FLAC to the phone and it plays them. However, I'm not sure if the app itself supports this sampling rate/bit depth natively or if it does any kind of on-the-fly processing to downconvert to a more standard format. I guess the case could be made that if I can't tell whether they are being played at full fidelity, it means that I don't need them in that fidelity to begin with. But humor me, if you will.
BTW, I'm talking about files that I copied to the device's microSD card, not files I uploaded to Google's cloud service.
Generate a 30khz tone and use a sound card line in running at 96 or 192 k to confirm its played back.
Generate a 30khz tone and use a sound card line in running at 96 or 192 k to confirm its played back.
Do you mean to play it through the line in and inspect the waveform visually to see if the tone is present, or was that a joke about how humans can't hear past 20khz?
Let's not forget that even if the phone can "play" 192 kHz files, there is no reason to expect the analog out to pass a 30 kHz tone.
Generate a 30khz tone and use a sound card line in running at 96 or 192 k to confirm its played back.
Do you mean to play it through the line in and inspect the waveform visually to see if the tone is present, or was that a joke about how humans can't hear past 20khz?
Completely serious. How else would you test this?
Generate a 30khz tone and use a sound card line in running at 96 or 192 k to confirm its played back.
Do you mean to play it through the line in and inspect the waveform visually to see if the tone is present, or was that a joke about how humans can't hear past 20khz?
Completely serious. How else would you test this?
Sorry for being defensive, I just know a lot of people around here are very anti-anything-above-16/44.1, and some can be a bit snarky. Thanks for the suggestion, I hadn't thought of it. I'll have to give it a try.
Any suggestions for tone generator software/site that will go up to 30khz? I'm finding a lot of online tone generators that stop at 22khz.
Here you are:
[attachment=7898:30kHz192-24.flac]
A 192kHz 24-bit stereo 10 second 30kHz tone.
It could possibly be the most exciting audio sample ever uploaded to HA
Cheers,
David.
Here you are:
[attachment=7898:30kHz192-24.flac]
A 192kHz 24-bit stereo 10 second 30kHz tone.
It could possibly be the most exciting audio sample ever uploaded to HA
Cheers,
David.
Haha, wow. Thank you! That was very nice of you. I'll report back on my findings when I get a chance to perform the experiment.
Any suggestions for tone generator software/site that will go up to 30khz? I'm finding a lot of online tone generators that stop at 22khz.
I've uploaded a test file with a sine sweep from 0 to 48 kHz. See the upload post (http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=105777&view=findpost&p=865893) with instructions.
Hope that helps.
Testfile sine sweep 0-48kHz (http://www.hydrogenaudio.org/forums/index.php?act=attach&type=post&id=7897)
Note: David was faster ! I should learn how to handle attachments
RMAA is also able to generate these test files. Just set it to 192kHz sampling rate.
Remember though, playing very loud ultrasound will tend to generate a lot of distortion. So will bad resampling. In many cases you will hear something when played through speakers. That does not necessarily mean anything. You'll need to use a high sampling rate ADC to figure out what is actually happening. RMAA is a great tool for this.
BTW, maybe I'm missing something, but what is the significance of the 30kHz tone if I'm trying to determine if it will play 192 kHz? Bear with me, I'm not the most knowledgeable person when it comes to the technical side of this stuff.
BTW, maybe I'm missing something, but what is the significance of the 30kHz tone if I'm trying to determine if it will play 192 kHz? Bear with me, I'm not the most knowledgeable person when it comes to the technical side of this stuff.
A 192 kHz sampling rate will theoretically allow signals up to 96 kHz to be represented. Most hardware will not remotely try to reproduce that, so a much more conservative frequency was suggested. If anything over 24 kHz is seen then that rules out, for example, resampling to 48 kHz.
An easy test is if you have a dog....they supposedly can hear up to 60KHz
BTW, maybe I'm missing something, but what is the significance of the 30kHz tone if I'm trying to determine if it will play 192 kHz? Bear with me, I'm not the most knowledgeable person when it comes to the technical side of this stuff.
The point of higher sampling rates is to enable higher frequencies. Hence, if you want to see if you're actually running above 48kHz, you'll need to play a tone above 48/2=24 KHz. 30 KHz is just a little higher, and so it should work at 96kHz (and higher) and fail at 48KHz (and lower).
Of course, you'll need a sound card that actually supports at least 96k to perform this test.
BTW, maybe I'm missing something, but what is the significance of the 30kHz tone if I'm trying to determine if it will play 192 kHz? Bear with me, I'm not the most knowledgeable person when it comes to the technical side of this stuff.
The point of higher sampling rates is to enable higher frequencies. Hence, if you want to see if you're actually running above 48kHz, you'll need to play a tone above 48/2=24 KHz. 30 KHz is just a little higher, and so it should work at 96kHz (and higher) and fail at 48KHz (and lower).
Of course, you'll need a sound card that actually supports at least 96k to perform this test.
Thanks for the info, I wasn't aware of the dividing by two thing. I have a current generation Mac mini, which supports up to 24/96, as shown on the Audio MIDI Setup utility included with OS X. Will report back later tonight with my findings. Thanks everyone for the help.
OK, I performed the experiment. Here are my results.
I copied the 30kHz test FLAC file kindly provided by David to my Galaxy S5 and ran a mini stereo cable from the phone's headphone jack to my Mac mini's line in. I made sure my sound card's input settings were set to 24/96. Using Audacity, I created a new track set to 32/192. When I began recording monitoring with the phone hooked up to the line in, I noticed a buzzing noise. However, when I started playing the test tone on the phone, the noise stopped, although the file was playing. When the file finished playing, the buzzing resumed.
I recorded several seconds of the "silence" while the file was playing and made sure there were no audible noises in the audio captured. I exported the audio to a 32-bit WAV file and then performed a spectrum analysis on the file using Audacity. Here's the graph that came up:
(http://s8.postimg.org/v5e5l74wl/Screenshot_2014_05_15_21_34_23.png)
So, does this prove that the Galaxy S5 can accurately playback 96kHz files at full fidelity, or did I do something wrong?
So, does this prove that the Galaxy S5 can accurately playback 96kHz files at full fidelity, or did I do something wrong?
I don't know about "full fidelity", but it does look like the software did not need to resample.
So, does this prove that the Galaxy S5 can accurately playback 96kHz files at full fidelity, or did I do something wrong?
I don't know about "full fidelity", but it does look like the software did not need to resample.
Great! I can rest well tonight with the knowledge that my phone's music player can play frequencies that are inaudible to human ears. I may have to rethink the need for high sampling rates. I'm not sure it justifies the amount of spaces the files take up. If you can't hear those frequencies, what's the point? Thanks again for the help, everyone.
Just so I don't leave anyone out, here is the frequency graph from Kees de Visser's 0-48 kHz sine wave sweep, as played back from my phone (thanks!):
(http://s2.postimg.org/tc7r1yz7d/Screenshot_2014_05_15_22_20_03.png)
What are you going to be playing these through? Headphones? IEMs? I've never seen any IEMs that go above 20KHz, I'm sure there are some but even very good ones don't go higher than this.
What are you going to be playing these through? Headphones? IEMs? I've never seen any IEMs that go above 20KHz, I'm sure there are some but even very good ones don't go higher than this.
And why would they? If you engineer something like that you're not doing your job. Unless they are made to fit animal ear canals.
And why would they? If you engineer something like that you're not doing your job. Unless they are made to fit animal ear canals.
I know My point was really about the space aspect too, 24/192 FLAC files are going to take up a lot of room on a phone, even if you want FLAC you could reduce that a lot by downsampling them all before copying.
Just so I don't leave anyone out, here is the frequency graph from Kees de Visser's 0-48 kHz sine wave sweep, as played back from my phone (thanks!):
It's a 0-96kHz sweep actually. When uploading the testfile I typed that wrong. Moderator Kohlrabi has corrected that, thanks!.
The frequency response you measured looks very good indeed. Are you sure the playback sampling rate was 192kHz ? The sweep frequency goes up to 96kHz but "only" 40-something comes out. This could indicate resampling to 96kHz. That's probably not enough reason to sell your S5 and buy a Pono (http://www.hydrogenaudio.org/forums/index.php?showtopic=94355) though
Lossless in general, and (IIRC) flac in particular, just isn't efficient for higher bitdepths - and, to a lesser extent, higher sample rates.
I'm not talking about audibility, or anything like that. What I mean is that the extra LSBs are effectively almost all noise, and hence unpredictable, and hence incompressible. For PCM, the raw datarate increases by 50% from 16-bits to 24-bits. For non-loudness war recordings, the increase in FLAC bitrate can be much greater.
The increased sampling rate isn't so much of a killer, especially on the many recordings where essentially "there's nothing extra up there" in the higher frequencies. However, I have a recollection (correct me if I'm wrong) that FLAC's predictors (and maybe its default black size) don't work as well at 192kHz as at 44.1kHz. Even if there's no extra real information (i.e. you simply upsample 44.1 > 192 essentially perfectly), the bitrate increases.
If you want measurable perfection beyond the bounds of the best human hearing while taming bitrates slightly, something like 20/96 with a gentle low pass filter starting at about 25kHz is a good choice. 24/96 or 24/48 run through lossyWAV with conservative settings (including, for examples, always keeping at least 16-bits intact) is another option. A higher bitdepth gives you a wider dynamic range (which you don't need) and the ability to turn the volume up during fades and reverb tails without hearing dither noise; a higher sample rate gives you a wider frequency response (which you don't need) and the ability to play back inaudible test signals instact and without intermodulation due to slow roll off DAC filters. It's things like this that are really important when you're listening to music on the move
Higher bitdepth and sample rate don't make things within the normal audible frequency and dynamic range more accurate. Certain people claim they do, but that's because they don't understand digital audio and, frankly, like to imagine things - or are trying to sell you something. People who believe this stuff often take great care with their recordings, and so they sound very good. The same recordings still sound very good converted to 16/44.1.
Cheers,
David.
The frequency response you measured looks very good indeed. Are you sure the playback sampling rate was 192kHz ? The sweep frequency goes up to 96kHz but "only" 40-something comes out. This could indicate resampling to 96kHz.
OP's sampling rate
on the recording side is known to be @ 96.
What are you going to be playing these through? Headphones? IEMs? I've never seen any IEMs that go above 20KHz, I'm sure there are some but even very good ones don't go higher than this.
I'll be playing them through relatively cheap earbuds, mostly. Guess that doesn't help the case for 96k.
Just so I don't leave anyone out, here is the frequency graph from Kees de Visser's 0-48 kHz sine wave sweep, as played back from my phone (thanks!):
It's a 0-96kHz sweep actually. When uploading the testfile I typed that wrong. Moderator Kohlrabi has corrected that, thanks!.
The frequency response you measured looks very good indeed. Are you sure the playback sampling rate was 192kHz ? The sweep frequency goes up to 96kHz but "only" 40-something comes out. This could indicate resampling to 96kHz. That's probably not enough reason to sell your S5 and buy a Pono (http://www.hydrogenaudio.org/forums/index.php?showtopic=94355) though
I'm not sure the playback sampling rate was 192, that's why I did the test. If it resamples to 96, I'm OK with that, as 192 is very much overkill for my purposes anyway (and 96 might be too). I'll stick with my phone, that Pono looks uncomfortable to put in your pocket.
I'll be playing them through relatively cheap earbuds, mostly. Guess that doesn't help the case for 96k.
So why do you want lossless 24/192 files again? I have good IEMs and listen to MP3s.
Lossless in general, and (IIRC) flac in particular, just isn't efficient for higher bitdepths - and, to a lesser extent, higher sample rates.
I'm not talking about audibility, or anything like that. What I mean is that the extra LSBs are effectively almost all noise, and hence unpredictable, and hence incompressible. For PCM, the raw datarate increases by 50% from 16-bits to 24-bits. For non-loudness war recordings, the increase in FLAC bitrate can be much greater.
The increased sampling rate isn't so much of a killer, especially on the many recordings where essentially "there's nothing extra up there" in the higher frequencies. However, I have a recollection (correct me if I'm wrong) that FLAC's predictors (and maybe its default black size) don't work as well at 192kHz as at 44.1kHz. Even if there's no extra real information (i.e. you simply upsample 44.1 > 192 essentially perfectly), the bitrate increases.
If you want measurable perfection beyond the bounds of the best human hearing while taming bitrates slightly, something like 20/96 with a gentle low pass filter starting at about 25kHz is a good choice. 24/96 or 24/48 run through lossyWAV with conservative settings (including, for examples, always keeping at least 16-bits intact) is another option. A higher bitdepth gives you a wider dynamic range (which you don't need) and the ability to turn the volume up during fades and reverb tails without hearing dither noise; a higher sample rate gives you a wider frequency response (which you don't need) and the ability to play back inaudible test signals instact and without intermodulation due to slow roll off DAC filters. It's things like this that are really important when you're listening to music on the move
Higher bitdepth and sample rate don't make things within the normal audible frequency and dynamic range more accurate. Certain people claim they do, but that's because they don't understand digital audio and, frankly, like to imagine things - or are trying to sell you something. People who believe this stuff often take great care with their recordings, and so they sound very good. The same recordings still sound very good converted to 16/44.1.
Cheers,
David.
Thank you, that's very good info. I guess the sensible thing to do is just do my own ABX testing and see if I can even tell the difference before devoting gigabytes to high res audio, which I haven't really done in a scientific way. I know 24/192 is complete overkill, but I'll have to do my own tests with 24/96. I don't really have a desire to do my own manual tweaking with going to 20/96 and using filters, since there are libraries of music already available in 24/96 (from HD Tracks, for example) and that is a pretty standard format , but I appreciate the suggestion.
The frequency response you measured looks very good indeed. Are you sure the playback sampling rate was 192kHz ? The sweep frequency goes up to 96kHz but "only" 40-something comes out. This could indicate resampling to 96kHz.
OP's sampling rate on the recording side is known to be @ 96.
Oh yeah, you're right! I totally forgot that the Mac mini's sound card is limited to 96k. That would explain why the recording frequencies were cut off for the 96k tone test. So, the phone can theoretically play back 192, but I have no way of verifying this because I don't have a sound card that supports that high of a bitrate.
Oh yeah, you're right! I totally forgot that the Mac mini's sound card is limited to 96k. That would explain why the recording frequencies were cut off for the 96k tone test. So, the phone can theoretically play back 192, but I have no way of verifying this because I don't have a sound card that supports that high of a bitrate.
Sample rate, not bit rate!
The frequency response you measured looks very good indeed. Are you sure the playback sampling rate was 192kHz ? The sweep frequency goes up to 96kHz but "only" 40-something comes out. This could indicate resampling to 96kHz.
OP's sampling rate on the recording side is known to be @ 96.
OK, if that's known, than it's fine. I just assumed 192kHz rate since the frequency axis of the posted FFT goes to 96kHz and the FFT applications I use automatically adjust the scale max. to the Nyquist frequency.
Thank you, that's very good info. I guess the sensible thing to do is just do my own ABX testing and see if I can even tell the difference before devoting gigabytes to high res audio, which I haven't really done in a scientific way. I know 24/192 is complete overkill, but I'll have to do my own tests with 24/96. I don't really have a desire to do my own manual tweaking with going to 20/96 and using filters, since there are libraries of music already available in 24/96 (from HD Tracks, for example) and that is a pretty standard format , but I appreciate the suggestion.
Seriously man. Don't pick an arbitrary floor with your ABX testing. After proving to yourself there is no need for anything beyond 16/44.1, see just how far down in bit rate you can go with lossy codecs. You may find that you're so grossly overestimating your needs that you're simply wasting tremendous amounts of space on your phone for no real gain. Almost my entire music library is FLAC, but I transcode it all to OGG Vorbis at around 96kbps to load on my phone because, seriously, on the vast majority of my music I just can't tell the difference between lossless and the q2 Vorbis.
For listening, 16/44.1 really is more than enough. And if you objectively test the modern lossy codecs you'll be quite shocked at how good they really are. I used to lossy encode, even AAC and Vorbis, at no less than 160kbps and thought I could tell the difference. ABX shattered that delusion, and ended up almost doubling the amount of music I can store on my devices!
I'll be playing them through relatively cheap earbuds, mostly. Guess that doesn't help the case for 96k.
So why do you want lossless 24/192 files again? I have good IEMs and listen to MP3s.
I have good headphones and listen to mp3s.
This thread is another example of pervasive audiophile nonsense causing someone to consider fixing something that isn't a problem (sample rate) before fixing something that is (transducers). People sometimes ask what harm can (inaudible) "improvements" do; here is your answer.
Cheers,
David.
P.S. I agree with yourlord entirely, though I use a higher lossy bitrate that I absolutely know I can't ABX (or haven't yet). That way, I can be 99.99% sure that any problems I hear not due to my encoding, but are on the originals, without even having to check. Every time I have checked, the mp3 artefacts I've thought I was hearing have been right there on the original CD (and usually not mp3 artefacts at all, though you'd be surprised what junk does get mastered to CD).
KitKat documentation clearly states that system provides downsampling without dithering. Regarding testing the files, it is wasting time. First at all you can't hear this frequency so you can't test using ears. Even if you get oscilloscope and connect to Samsung phone jack, you can still see nothing, just because Samsung audio analog track is limited to 20KHz.
If you want to get full advantage of 24/192 you need to connect an external UBS DAC and make sure your audio program is capable to use it. Not so many audio players can utilize it now, but some will be soon. I am holding buying S5 until Kamerton will support external DACs.
If you want to get full advantage of 24/192
And that advantage would be...?
@MOCKBA: Do you realize that the same post that you are replying to contradicts what you are saying?
KitKat documentation clearly states that system provides downsampling without dithering. Regarding testing the files, it is wasting time. First at all you can't hear this frequency so you can't test using ears. Even if you get oscilloscope and connect to Samsung phone jack, you can still see nothing, just because Samsung audio analog track is limited to 20KHz.
The only waste of time was your reply from ignorance. Playback of audio well over 20khz has already been demonstrated.
Even if you get oscilloscope and connect to Samsung phone jack, you can still see nothing, just because Samsung audio analog track is limited to 20KHz.
FWIW, this is only true at 44.1kHz. On a modern oversampling DAC (which is the type used in smartphones and other portable electronics), the reconstruction filter cutoff actually scales proportionally with the sampling rate. So it'll be around 20kHz for 44.1KHz, 22 KHz for 48k, 43KHz for 96k, and so on.
This is because the actual analog filter is set at about 1/4 to 1/2 the the oversampled rate (typically 5-10MHz) and a variable digital filter is used below that. Furthermore, when you change the sampling rate, generally the oversampling ratio is adjusted to maintain a nearly constant actual sampling rate. So for example, at 48kHz, the player might oversample 256x for a rate of 12.3MHz. At 96k the oversampling ratio will be changed to 128x for a rate that is still 12.3 MHz. However, the bandwidth of the digital filter will be fixed at approximately 90-95% of the Nyquist rate, hence the actual bandwidth scales with sampling rate, and SNR is approximately constant.