|
Post by popmann on Mar 13, 2018 13:57:55 GMT -6
Ok, but I'm using bone standard SATA spinning drives too. No issues here with speed, no SSD needed to get this performance. I am too, svart. On Windows...the 2012 Mac, is SSD from the factory, so I can't speak to that...but....I've recent reformatted the external USB3 SSD I bought for VIs on the Mac NTFS for use on the old box, as it's just downright BETTER at low latency VI playing. He doesn't need the SSD for the HD audio. He needs it for running his VIs AND streaming HD audio from the same drive. A sample steaming VI....is literally THE benchmark for drive performance along with video editing scratch disks--but those actually bench two different aspects of performance. Since it apparently wasn't clear--he can produce 24/88 audio full productions with not only his current system AS IS....but, also his last system with the i5.
|
|
|
Post by Bob Olhsson on Mar 14, 2018 11:11:13 GMT -6
Nyquist specified infinite slope filters with zero ringing. That doesn't exist in the real world.
Research in the 1980s determined that the optimum real-world sample rate to keep crap out of the audible range below 20 kHz. is 60kHz. The Society of Motion Picture and Televisions engineers fought Sony and Panasonic to get 48kHz. x 20 bits as their minimum standard.
|
|
ericn
Temp
Balance Engineer
Posts: 15,011
Member is Online
|
Post by ericn on Mar 14, 2018 12:11:10 GMT -6
Nyquist specified infinite slope filters with zero ringing. That doesn't exist in the real world. Research in the 1980s determined that the optimum real-world sample rate to keep crap out of the audible range below 20 kHz. is 60kHz. The Society of Motion Picture and Televisions engineers fought Sony and Panasonic to get 48kHz. x 20 bits as their minimum standard. I believe you meant Sony & Phillips, but Panasonic and everybody else in consumer electronics and all the labels adopted the standard ( the true wonder of CD was the universal adoption of the standard) !
|
|
|
Post by jazznoise on Mar 14, 2018 13:37:04 GMT -6
Just use 48khz and be done with it. At least with the setup I use I find it's the best tradeoff, I didn't expect it to be better than 44.1 but it is.
|
|
|
Post by Bob Olhsson on Mar 14, 2018 14:28:46 GMT -6
I was at some of the meetings. It was Sony and Panasonic. You are thinking CD.
The CD standard was adopted quickly because it used the Phillips video disk that could be replicated in all vinyl plants. (Can you say PCM-F1?)
|
|
|
Post by Martin John Butler on Mar 14, 2018 15:29:44 GMT -6
jazz noise said, " I didn't expect it to be better than 44.1 but it is".
Same here.
|
|
|
Post by gouge on Mar 14, 2018 16:01:04 GMT -6
yup, and 96 is betterer again in my experience. 96 is the minimum i'll record at.
|
|
ericn
Temp
Balance Engineer
Posts: 15,011
Member is Online
|
Post by ericn on Mar 14, 2018 17:50:34 GMT -6
I was at some of the meetings. It was Sony and Panasonic. You are thinking CD. The CD standard was adopted quickly because it used the Phillips video disk that could be replicated in all vinyl plants. (Can you say PCM-F1?) I didn’t know they were in that early, the guys from broadcast, who made all the Ramsa and DAT machines always played it like they were just following Sony and Phillips, but that’s why your the beard of audio knowledge, you were there!
|
|
|
Post by Bob Olhsson on Mar 14, 2018 20:15:06 GMT -6
This was a standard for digital audio tracks on video recorders and digital television. It was after the CD.
|
|
|
Post by drsax on Mar 15, 2018 9:24:17 GMT -6
My experience is not about which sample rate sounds best inherently, but rather, about how a song sounds after it’s finished and mixed and mastered and converted to a lesser sample rate for current music distribution. I’ve found that I’m happier with a great sounding 24-bit 44.1kHz project that stayed at that sample rate from start to finish than a project at a higher sample rate converted later. It would be nice if high res audio would take a hold at a consumer level, but it hasn’t yet. I also have heard many projects at 44.1khz done right that sonically hold up to material done at higher sample rates. Save for sparse classical or maybe really exposed jazz or folk, I prefer staying at 44.1kHz
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Mar 15, 2018 10:12:55 GMT -6
This has also been my way of working for a few years now, stay at the native sample rate until the end. I did do tests back then and this is what I thought sounded better. However, as mentioned above I'm interested in the "other" way again too, and so just for shits and giggles for this current project (19 track compilation with many different formats and sample rates submitted), I have decided to upsample everything to 96 and do all the work that way, making sure all OS in any plugins along the way is turned off (if it's switchable). Mastered five of the tracks so far, and am liking what I am hearing, after mastering and conversion back down to 24/44, BUT there's definitely a bit more work involved. That's fine if the results are better though. If Bob O. does it and swears by it I wanted to give it another try.
|
|
|
Post by Bob Olhsson on Mar 15, 2018 14:11:15 GMT -6
My tests were all comparing final 44.1 files.
I was using Sonnox, DMG, Sonoris, SoundToys, PSP, and Waves upsampled to 96. Interestingly they didn't sound as good upsampled to 88.2. I have a hunch the developers did their listening tests for code simplification at 48 or 96. The TDR plug-ins sounded better at 44.1 but not always as good as the others at 96.
The best was when original audio had been recorded to 96k in the first place. When I realized that, I bit the bullet and bought a new computer.
|
|
|
Post by gouge on Mar 15, 2018 16:09:43 GMT -6
If I had the coin id record dsd.
Love the fidelity.
|
|
mhep
Full Member
Posts: 36
|
Post by mhep on Mar 17, 2018 18:43:00 GMT -6
Just a couple last points.
96kHz latency benefit is indeed worth the change for anyone working natively.
Circling back around to fallacy that more sample points equals a more accurate waveform—that would mean that even recording at an 8kHz samplerate (yes, 1/12 of 96kHz) would have MUCH more accuracy (more sample points) at 40Hz than 96kHz has at 4kHz.
It's pretty simple. More than two sampling points for a cycle means nothing.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Mar 26, 2018 3:15:04 GMT -6
Actually, i even upsample 48kHz recordings to 96kHz, i guess i still profit from several advantages, like e.g. lower latency when adding VIs, staying 96kHz throughout the following processing, so i can skip some upsampling in specific plugins (like B. Ohlsson pointed out) etc. ... My machines are actually pretty much equal to Martin's Mac, and i am using Mixbus, which has already a much higher processing load due to the analog console emulation throughout the audio path. No problem working in 96kHz for me. If i run out of processing power with excessive high quality plugin usage, i freeze tracks and that's it (and despite the fact, that this is a bit more work, because right now, Mixbus does not have a comfortable track freeze function, it's worth it IMHO). I can only encourage to watch out, what the bottleneck is in a specific system, if you encounter problems in 96kHz. Could be driver issues with south bridge, harddisk, processes that interfere with the audio system etc. etc.. In the end you are at least sure, that you have a fully functional system. Software like DPCLat under Windows can help investigating at least some of the issues. I do not look back to 48kHz anymore, at least for downmix and mastering duties. Even if i import from 48kHz, it is only one time up- and downsampling, while there is a lot of processing inbetween. Using today's good src algorithms i think it is not a problem for me at all.
|
|
|
Post by viciousbliss on Mar 26, 2018 6:50:17 GMT -6
I've read stuff that says SRC can be done accurately. There's that one site that shows the reconstructions from various programs. Audacity always seems to be the best. I definitely hear big changes when converting an 88 or 96 24-bit file to 16 and either 44 or 48. Same if I turn the 16-bit dither on in either Peacock or Oxford Limiter. I'll have to try converting an 88 or 96 to 44/ or 48 and leave it all at 24 bits. The worst part about 88k or above is cpu. With Azure's latest update, I find myself doing a lot of guessing which settings to change and hitting bounce. Sometimes I can get real-time playback, but not too often. The HQ version never gets real-time playback even if it's the only plugin active. I will have to try it on my erratic Xeon workstation which scored around 1100 multi-core points on cpuuserbenchmark. Single core was bad at around 70, but that doesn't seem to matter anywhere as much as the multi. That 1100 is almost twice the 600 this computer scores. Maybe a minimum setup for a 30+ track session with one or two cpu-intensive plugs should include either a Ryzen 1600 or Intel 6850k, both scoring around 1000 on multi. These years old Intel chips seem to be rising in value ironically. One would think if a Threadripper 1950x is around $700 that an Intel 6950x would not still command prices around $1500 with stock selling out. AMD prices continue to fall as the 1600 was $149 the other day. Ryzen 2 will be out soon too.
The type of plugin matters a lot. From what I recall from the ultimate plugin analysis discussion, pure digital plugins are fine at 44/48 in terms of aliasing. Other stuff like EQ cramping could still happen without the plugin doing something about it.
|
|
|
Post by svart on Mar 26, 2018 9:29:03 GMT -6
Just a couple last points. 96kHz latency benefit is indeed worth the change for anyone working natively. Circling back around to fallacy that more sample points equals a more accurate waveform—that would mean that even recording at an 8kHz samplerate (yes, 1/12 of 96kHz) would have MUCH more accuracy (more sample points) at 40Hz than 96kHz has at 4kHz. It's pretty simple. More than two sampling points for a cycle means nothing. Well, I work in high-speed digitization (RF capture) and that's entirely wrong. Two points per cycle is the minimum for recreation of a sinusoidal waveform at a specific frequency that's related to the sampling speed. However, if you want/need to see the tiny pertubations(noisy inclusions on the waveforms) you MUST have much higher precision, AKA: higher sampling rates for the same specific frequency. Those tiny noise inclusions mathematically equate to "high frequency" noise, that would be normally filtered by the anti-aliasing filter and sampling rate. Just because you think the fundamental (or in this case, carrier) frequency is well within the sampling band, doesn't mean you don't want to see what's being modulated onto it in finer detail.
|
|
|
Post by Bob Olhsson on Mar 26, 2018 13:53:59 GMT -6
Indeed and that minimum assumes perfect, infinite slope filters with no ringing which don't exist in the real world.
|
|
|
Post by swurveman on Mar 26, 2018 15:38:47 GMT -6
Just use 48khz and be done with it. At least with the setup I use I find it's the best tradeoff, I didn't expect it to be better than 44.1 but it is. I will say this: I record and mix at 48khz and Wavelab's Crystal Resampler sounds better (to my ears) than Cubase's resampler when I go down to 44.1khz. So, whatever you do, make sure you're downsampling with something you like.
|
|
ericn
Temp
Balance Engineer
Posts: 15,011
Member is Online
|
Post by ericn on Mar 26, 2018 16:53:55 GMT -6
Just use 48khz and be done with it. At least with the setup I use I find it's the best tradeoff, I didn't expect it to be better than 44.1 but it is. I will say this: I record and mix at 48khz and Wavelab's Crystal Resampler sounds better (to my ears) than Cubase's resampler when I go down to 44.1khz. So, whatever you do, make sure you're downsampling with something you like. I know mastering engineers who choose SRC on a case by case basis!
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Mar 27, 2018 1:42:06 GMT -6
I will say this: I record and mix at 48khz and Wavelab's Crystal Resampler sounds better (to my ears) than Cubase's resampler when I go down to 44.1khz. So, whatever you do, make sure you're downsampling with something you like. I know mastering engineers who choose SRC on a case by case basis! I do, but it's a speed vs. sound quality thing. If I want the best sound and have the time, I use FinalCD, if I'm in a hurry then iZotope, but TBH there's not a huge amount of difference.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Mar 29, 2018 7:09:21 GMT -6
Here's a bit of a screen shot from iZotope RX. It's a basic spectrogram, so the brightness of the signal indicates strength. The graph covers about 3 minutes. The signal is from an AEA N8 ribbon, one of the mics I used on a concert recording a couple of days ago. I recorded at 192K. It's a good example of signal that's present when people insist there can't be. Frequency is along the right side of the image. The N8 begins to roll off a little above 4K and you can see that in the graph (about a third of the way up). But what should interest people is the fact that there's signal up to 60 kilohertz. I could raise the sensitivity of the graph and show there's signal above that. And remember this is from a pretty classic ribbon--one that people use because it's a little 'slower'. Just because a microphone spec doesn't give numbers above 20K doesn't mean there's no signal there. For the record, the source is a string quartet. Nothing with exaggerated HF content. Take a look at the area above 20K (the top third of the image). There's still energy up there. What would happen at a 44.1 sample rate? Without the mythical perfect filter, some of that energy will reflect back down below 22K. This is what Bob is talking about when he says the primary difference between low and high sample rates actually appears in the midrange. If that energy is recorded to disk, it will pass through every EQ, every compressor, every reverb, every single gain adjustment. This is one of the advantages of recording at high sample rates--even when the deliverable is 44.1/48K. You'll only do a single downsampling pass. Depending on the quality of the downsampling logic, you'll still get some of that folded-over signal. But you'll only get it once. Your source tracks will be cleaner.
|
|
|
Post by Martin John Butler on Mar 29, 2018 7:41:27 GMT -6
Well, there it is. I've always thought that even though I'm no engineer or mathematician, there's something in the air when you compare higher sample rates. Of course every time you have these conversations some knucklehead insists you prove your point, as if I have nothing else to do than research scientific audio test data to try winning an argument.
My ears are my proof.
|
|
|
Post by svart on Mar 29, 2018 10:43:11 GMT -6
Well, there it is. I've always thought that even though I'm no engineer or mathematician, there's something in the air when you compare higher sample rates. Of course every time you have these conversations some knucklehead insists you prove your point, as if I have nothing else to do than research scientific audio test data to try winning an argument. My ears are my proof. I don't understand what you're arguing here.. That graph was 192k, which has usable Nyquist of 96k.. I don't see anything above roughly 70k on that graph, AKA nothing unexpected.
|
|
|
Post by svart on Mar 29, 2018 10:47:49 GMT -6
Here's a bit of a screen shot from iZotope RX. It's a basic spectrogram, so the brightness of the signal indicates strength. The graph covers about 3 minutes. The signal is from an AEA N8 ribbon, one of the mics I used on a concert recording a couple of days ago. I recorded at 192K. It's a good example of signal that's present when people insist there can't be. Frequency is along the right side of the image. The N8 begins to roll off a little above 4K and you can see that in the graph (about a third of the way up). But what should interest people is the fact that there's signal up to 60 kilohertz. I could raise the sensitivity of the graph and show there's signal above that. And remember this is from a pretty classic ribbon--one that people use because it's a little 'slower'. Just because a microphone spec doesn't give numbers above 20K doesn't mean there's no signal there. For the record, the source is a string quartet. Nothing with exaggerated HF content. Take a look at the area above 20K (the top third of the image). There's still energy up there. What would happen at a 44.1 sample rate? Without the mythical perfect filter, some of that energy will reflect back down below 22K. This is what Bob is talking about when he says the primary difference between low and high sample rates actually appears in the midrange. If that energy is recorded to disk, it will pass through every EQ, every compressor, every reverb, every single gain adjustment. This is one of the advantages of recording at high sample rates--even when the deliverable is 44.1/48K. You'll only do a single downsampling pass. Depending on the quality of the downsampling logic, you'll still get some of that folded-over signal. But you'll only get it once. Your source tracks will be cleaner. Nyquist folding would mathematically be unity gain, but the folding would put the mixed signal inverted, and placed within the reject slope of the analog filter, doubling the reduction in amplitude. This also doesn't account for the rolloff due to the A/D input structures that limit the bandwidth. So any folding would be much less pronounced than you'd expect, and likely leave the mid-range rather clean.
|
|