|
24/192 Music Downloads ...and why they make no sense
|
View this Thread in Original format
| Psyshell |
I already knew some of this, like the fact that vinyl in no way sounds better than cd (not that that's specifically mentioned) but it never occured to me that 192khz music can in fact sound worse.
There is one thing the guy didn't delve into though. Sure if you're having supersaws then perhaps having the 44hz and 22hz frequencies in it isn't good... but, if it's the sound of a guitar recorded, who's to say that adding those frequencies in is actually making the guitar sound worse? Just because when you only hear the 10khz artifacts from the frequencies intermodulating and that ends up sounding a bit weird/like noise doesn't mean that it is in fact going to sound worse when mixed in with the rest of the guitar sound. It never occured to me just how big the dynamic range was. I guess I had it in my head that it was perhaps 20db lower or more than what it is. The quote "cds were made to capture the full range from one orchestra to a single instrument" is wildly inaccurate. Their dynamic range is far higher than that. It's kind've depressing in a way that there really is nothing better than cd audio but oh well!
The one gap in the subject that I currently have is I do have to wonder how scaling from 44khz to 48 or visa versa doesn't make an audio signal sound worse. If it was pixels in a movie it'd make it look far worse because they don't convert between the two well. Surely converting from one to another would involve adding in additional sine waves and would just cause an in-approximate mess.
I do however stand by that 320kbps is fairly obviously distinguishable from lossless though. Especially if you've heard the source in lossless before. |
|
|
| echosystm |
| quote: | | Originally posted by Psyshell The one gap in the subject that I currently have is I do have to wonder how scaling from 44khz to 48 or visa versa doesn't make an audio signal sound worse. If it was pixels in a movie it'd make it look far worse because they don't convert between the two well. Surely converting from one to another would involve adding in additional sine waves and would just cause an in-approximate mess. |
Pretty much. Upsampling is always very dirty unless the target rate is a multiple of the original rate (you can make interpolation good times in that case). |
|
|
| Looney4Clooney |
| Nyquist theorem is not exclusive to audio. Saying it isn't good enough for audio makes no sense. Upsampling makes a lot of sense. But given your use of a term that is pretty basic and important in digital audio, I am pretty sure the basic concept is complete gibberish to you. |
|
|
| echosystm |
| quote: | Originally posted by clay
nyquist theorem is that if you have the double sample rate of what you plan to sample it should be enough to recreate it. |
Wrong.
The whole point of Nyquist is merely that you don't lose the frequency entirely. No more, no less. It doesn't try to make any guarantees regarding your perception of "quality".
| quote: | Originally posted by clay
however if you sample a sinus at 22kHz using a 44,1khz samplerate it wont be a sinus but trangle wave, or maybe even worse a square pulse wave. |
Anything above 20khz will be high passed to avoid aliasing, so trying to sample 22khz with a 44.1khz sample rate is retarded. Why do you think we use 44.1khz instead of 40khz? The whole point is that the extra 4.1khz gives you a ~2khz transition band to cut out frequencies that can't be sampled accurately. |
|
|
| Looney4Clooney |
Nope, response to clay , not echo boy , quote function too much work
it is a math model that assumes certain things that don't occur. But n your instance, well to start, it isn't double but rather the rate sampled must be. Less than half. It doesn't state an actual rate for audio, merely a method to achieve reproduction which does work assuming certain things that don't occur in real life ie perfect band limiting.
You always seem to focus on the wrong thing. Like every ing discussion, you focus on what isn't the issue.
If anything could be said about sound quality and it's limitations, it would be the analog playback. Speakers really haven't changed and until we I find a way that delivers sound how nature intended , it will always. Sound like a recording. |
|
|
| MaxC |
| quote: | Originally posted by echosystm
Pretty much. Upsampling is always very dirty unless the target rate is a multiple of the original rate (you can make interpolation good times in that case). |
While this may have been the case with naive, rudimentary implementations, my understanding is that this is no longer the case. One can always scale up to the least common multiple of the source rate and target rate and divide back down to the target rate without introducing any additional artifacts beyond those inherent to the upsampling process. This comes at additional CPU cost, granted, but that is irrelevant if quality control is your priority. I looked into this a great deal when trying to decide whether it made more sense to upsample from 44.1 to 88.2 or 96 khz. Dan Lavry discusses it some in a whitepaper on SRC if I recall correctly. |
|
|
| echosystm |
| quote: | Originally posted by MaxC
While this may have been the case with naive, rudimentary implementations, my understanding is that this is no longer the case. One can always scale up to the least common multiple of the source rate and target rate and divide back down to the target rate without introducing any additional artifacts beyond those inherent to the upsampling process. |
That's like saying "i'll punch you in the face, but it's ok because you won't feel any pain other than being punched in the face". :stongue:
I'm happy to be proven wrong, but all you've done is outline one of the two approaches to sample rate conversion (the other is interpolation as a time series). In either case, rounding is going to happen unless the target rate is a multiple of the original rate. |
|
|
| Looney4Clooney |
| It's just math. They both result in the same end. One requires extra operations. |
|
|
| DJ RANN |
There's a lot of bull in this post and although he's theoretically correct in the argument that the extra headroom is of no consequence due to our threshold of hearing.
However, what he doesn't mention is that in a professional environment and with pro level equipment, the actual real world dynamic range of 16bit systems is 93db and for 24bit it's around 116db - that extra 21db of headroom is actually useful on a calibrated system, especially when doing any recordings.
It's also useful due to the fact the noise floor is inherently proportional to the signal - with a higher bit depth and therefore usable headroom you can jack much quieter sounds up without that noise floor being so much of a problem.
One of the people in the thread state some interesting things such as there is no such thing as 24bit DAC output, only 20bit as resistors and capacitors are a bottle neck to the overall output range, and if this is true, only using a 16bit output stage would result in only about 12bits of headroom which is now getting dangerously low in terms of not having a enough dynamic range for the human thresholds of hearing as they relate to headroom.
I've stated on here plenty of times that analogue tape only has a usable bit depth of 60db but people love it because of how it sounds, and that is also the main flaw with the CD vs Vinyl debate.
Looking purely at bit depth, CD wins, but when you have to look at sample rate with the same argument, vinyl has no sample truncation or aliasing so it is the clear winner in terms of audio reproduction.
He also glosses over three other major points:
1, Why did Steve jobs NOT want 24bit recordings? It was not a technical discussion of bit depths. It was purely the ratio of Money per Megabyte. an Apple lossless 24bit file of 5 mins = 100mb. A 320kbs MP3 of the same length? 10mb at most.
For the great unwashed who will never use anything more than ipod or beats by dre headphones, MP3 Is all they will ever know so there's no point apple using 10 times the storage space for billions of downloads when it won't make a shade of difference to 99.9999% of their users.
2, I'm calling bull on the author regarding his claim to be able to identify each MP3 encoder based on it's sonic qualities. While I'll concede back in the early 90's there were some very rudimentary encoders which at low bit depths could easily be told apart due to how badly they introduced articles and coloured the sound, there is no ing way you could do this on even slightly higher bit depths with anything more recent than say 1998 technology. I've worked with some of the best pro engineers in the world and while some of them can tell you that something sounds different to another identical recording, they can't blindly identify encoders to any accuracy, even though they work day in, day out with this technology and have done for 30 odd years.
This guy, quite literally would have to be the only guy in the world who has ever been able to do this.
3, The final problem he glosses over is to work, dither introduces random white noise to essentially spread the errors so we don't hear them. Whether you like it or not, it's a form of pyschoacoustic masking. We may not perceive it outright but it's there, and within the human threshold of hearing. There is again an advantage (that someone points out later in the thread and the author has to concede it) that having the lower relative noise floor in 24bit (over 16bit) is an advantage of of the higher bit depth.
Will any of us ever be able to hear that difference? Doubtful, but it's still there. |
|
|
| echosystm |
| quote: | Originally posted by clay
I still am unsure due to the nyquist theorem possible faults. |
| quote: | Originally posted by clay
interestingly enough it seems that 44,1kHz was not chosen due to the space required for a transition band |
You missed the point entirely. The first point was that your example of trying to sample 22khz is retarded, because it will get filtered anyway. The second point was that we use rates higher than 40khz (eg. 44.1khz), because they provide space to do the aforementioned filtering. There's no "fault" with Nyquist. . |
|
|
| MSZ |
| This thread has been palm'd. |
|
|
|
|
