Great video, in-depth explanations, solid. I record at 48kHz if I record for music production. For foley and sound design experiments, I always record at 96kHz - it's always nice to able explore the otherwise inaudible material of especially ambient recordings. So, 48 kHz for when you know what you want on the recording, and 96kHz when you want to explore.
You are pretty close, but for me the benefit of working at 96 is that anti aliasing filters on my interface don’t give me high end phase shift in the audible range, but at 44.1 and 48 they do and I don’t like the harshness. It’s most noticeable on things like hihats, cymbals, clean electric guitar, and vocals.
When I first used 96k with a real drummer, the cymbals sounded close to analog tape again. I just used a 2” tape machine and it was down 3dB at 30kHz. You need source material that shows that difference to hear it. I think 96k has better phase and group delay characteristics. Most instruments do not have harmonics above 10k to make a difference.
I don't know about you guys.. But I noticed that working at 96khz gives next pros: Synths calculated more accurately, its sounds more detailed and pronounced. Reverb has more open feel, also giving closer to life result. Delay has more details, sounds repeats more clearly. And you always can transpose 96khz down without losing highend. Cons: it's takes more space and eats more CPU.
Using an older workstation with a 96khz sample rate and I can tell that the detail of the high end is easier to follow when I'm mixing as if I can hear where the sound is coming from in the mix better, even for dirtier mixes. Maybe it's placebo but idk
@@ELISHACAEZ I think it's big advantage, even aliasing will be dramatically lower. I always getting like "real life sound" while working and writing in 96khz. All effects seems to give more quality than they can at 48khz for example.
hi, sorry but i'm very new to this and searched a ton, but i couldn't find something that could help in my situation i want a battery shotgun mic that could get that 96Khz, is it even has anything to do with the mic?? could i get any mic that i want and connect it to a pc and record at 96khz?? i would really appreciate if you could help, thanks a lot
@@ali32 Allow me to help. Yes any mic that you use can be recorded at 96khz sample rate, and will sound better as opposed to say 44.1khz. Also any software plugins or Virtual Synths you may be using if on a DAW on PC will also sound clearer but takes double the CPU power and also disk space for storage. I hope this helps mate.
Absolutely killer vid my dude. You brought new information on a discussion that has been beaten to death (most people just repeat other people's info). Good job with this example. Keep it going.
I think 48 is the best balance and also makes computations for oversampling easier on your CPU. 48 is also the industry standard for audio embedded in video for streaming, DVD, and BluRay. If you're not doing time stretching, 48 is fine. If you're doing time stretching, sure, go for 96 or 192. But as you said, you could just up sample the one audio clip, so...
Video is definitely pushing it in this direction. It's also nice to have a bigger "window" to reduce aliasing and the effects of the anti-aliasing filter. Cheers!
I record and mix at 48kHz ,while my music buddy does everything at 44.1 and he sends stuff as MP3s. Both of those drive me crazy, especially because I render my stems at 48kHz 24 bit and I know it's going to convert at his end into 44.1. If he send me stuff back it's going to be upconverted to 48kHz again. And MP3s just mess with the bottom and top end. Smeared lows and dull highs.
You might want to add: MP3 does mess with it no matter if it’s 44.1 or 48 kHz, due to its compression, which is quite lossy. Converting 44.1 to 48 kHz does not change anything in sound, unless ofc you let the windows mixer do the job. You can not regain missing frequencies and the ones that are there at 44.1 kHz will sound the same, no matter how many samples you add to match whatever desired sampling rate.
I'm coming from a FOH perspective. There's a trend toward larger sample rates in new digital FOH systems. Most of the people having discussions of sample rates are among recording engineers, so I'm looking for more answers, particularly as it applies to live production. The aliasing makes sense. If you are summing two signals at the same frequency that are fraction of a sample out of phase, you can theoretically end up with a sum that has an artificial frequency within the audible range. This is especially true if you do parallel processing on a signal that doesn't have good time alignment correction in its processing.
@@shoepedals I'm talking about sum and cancellation in very high frequencies that in a lower sample rate results in a wave shape or synthetic frequency in the audible harmonic structure that alters the tone of the sound, particularly if the system has multiple resamples in the signal chain (Digital wireless mic with analog out to digital preamps for a console that has analog outs to the DSP that has analog outs to the amps.). A higher sample rate won't have this problem.
Thanks for the detailed video and great research behind the scenes! Just confirming - indeed, higher sample rates make flex time editing in Pro Tools so much better. I recently recorded metalcore drums at 88.2 kHz (with no real reason), and after heavy editing, everyone involved in the project was blown away. Much fewer editing artefacts and much more transparent result.
That's awesome! I haven't heard anyone else ever mentioning this, so I'm glad to hear you had a similar experience. Everything was pretty apparent that the higher sample rates are way better for editing. Thanks for sharing your experience!
@@RaytownProductions I wasn't expecting this and that was an accident, so now your video proves the concept! I'll aim to record at higher sample rates where I can. Some outboard gear in the studio only allows 48 so that's a limitation. If very heavy editing is needed, I'll disregard that gear now, after knowing this principle
@@DrMikeMetal good news! This trick works even if you record at lower sample rates! Just upsample then edit and then down sample back to your native rate. Best of both worlds!
That's one of the main reasons I use higher sample rates. I will often do a lot of tempo stretching early on with the bed tracks to get certain variations in. Any time I'm doing this kind of work on a project, higher sample rates are a lifesaver.
My comment may be trivial- but a few years ago I recorded a track as a warm up while awaiting for other band mates to arrive. The engineer recorded me playing bass (Chameleon-Herbie Hancock) at 384 sample rate. I couldn’t even believe the difference - it was perfect. I was sold thennn with the exception that most folks record at a lower rate.
One more benefit of 96 Khz over 48 and 44,1 - lower latency during recording. Buffer size is always in samples, not ms. That means that buffer size of 96 samples in 96 KHz is lower than in 44.1. That's more convenient for guitar recording.
I dont understand this because when i record and mix at 96khz, the mixes sound cleaner, more open, and wider stereo. You and many say the same things the higher sample rates does not improve quality, but my ears tell me the contrary.
Try a double blind test. There is a software called abx to do blind shootouts. If you know which is which going in, you might have an unconscious bias. This is the only way to know for sure. In blind shootouts for the music I mix, I cannot statistically tell a difference. Hope that helps!
I record at 96KHz because i have much better detail at high frequencies (above 10KHz) especially on drums. You can also try bone conduction (wired - piezoelectric as bone conduction) and you able perceive around 50KHz (the vibration goes directly to cochlear through skull // bypasses the eardrums) but you need increase 30dB above 20KHz in EQ to start hearing - offcourse it depends if some people are able to hearing with this method.
I have been all over the place as far as what sample rate to use for tracking and mixing. Most of the time, I use 44.1 as it gets the job done for what I normally do and it's often going to end up that way in the end. For instance, I have learned that when cutting records from a digital source, it's at 44.1. I have used 48 and I may use that more as that is what is used for video. I'm kinda miffed that 88.2 is not a more common option - you have to get a high-end field recording device to have it as an option - because it is cleaner to downsample from there to 44.1 by half compared to 0.459375 from 96. I would say that the only reason why you would want to *record* at a higher sample rate from the start is if you have certain analogue gear with some superharmonic colour you would like to see printed into the digital realm.
Same! I typically go with 48 khz so it's compatible with video but I can't really hear much of a difference between 44/48. But with editing, man it is so much better at 96 or even 192!
Yes, but you are bouncing with aliasing and artifacts from internal oversampling. Some plugin companies either cannot or will not even say to what degree their products oversample.
Great video. A couple years ago I asked on a forum if there was a benefit to recording at 96kHz if time-adjusting. I got some great insights into 44/48 vs 96kHz but no answered about stretching audio. So thanks! Great to have that question answered!
I record at 48kHz because I hear more "air" in the audio. I know this shouldn't be the case, in theory, but I think it has to do with the filters - higher sample rates have higher filters so less air is removed at the top end.
I think you are close but it’s the opposite, the filters at 96 are high enough where they have no effect on the audible range, where as at 44.1 or 48 it sounds a little brighter because the filters are closer to the audible range and steep enough to cause resonance in the upper audible range. So 48 is adding high end from the bump of the anti aliasing filters.
@@MuddyRainMusic ideally the roll-off of an anti-alias filter starts from 0dB and heads towards -inf. dB, so there really should be no bump at the cutoff frequency when it comes to frequency response. Instead, the resonance you're talking about is caused by the steepness of the roll-off. Perhaps ringing would be a better word to describe what's actually happening, since the cutoff frequency "rings" for longer after a steep filter cut. You can see and hear it for example by bandpassing a pure sine wave at the exact same frequency the wave's at, with a very high Q (and a typical constant-peak bandpass filter to make sure the frequency response is at 0dB where the sine wave is). Turn the wave on/off without fade ins and outs, and you should notice it takes the length of the filter's impulse response to actually change between on/off state, instead of changing instantly as it was before the filter.
Note: If I'm not mistaken, some Universal Audio plugins only work with oversampling above 96k. Working with high sampling can be advantageous to reduce latency.
Artifacts from internal oversampling of plugins exist. I sometimes use a difference mix session for the element in question against a stereo bounce of everything else = no aliasing or artifacts. I bounce each stem out of their respective 96k sessions AT 96k so that the compression and basic eq & saturation are "baked in" without those aforementioned issues. I then convert THOSE bounces to new 48k bounces which just downsamples it enough to compile all of the together into one cpu friendly session for a full mix. Any further plugins that might cause a bit of aliasing are usually minimal at that point. Volume automations, some ducking, and it's good. Just to be safe, I usually only use roughly 2 reverbs on each production. Those reverbs/settings are mirrored across all session files for that song. I will sometimes commit the reverb to 96k stems as well. Yes, I often have to make revisions and further bounces from the 96k stem sessions. ...all because I can't afford 100k in hardware haha
Great video. I bought a DAC that can easily do 96khz but no matter what I listen to, and no matter from what device, it is ALWAYS 48khz. What's my problem?
Hi...This was really helpful...thanks! So, I have started on GarageBand on an iPad pro in recent years, doing my vocals on a 16 bit/48 (44?) khz mic...It records great, but, has a very small screen. I'm a trained vocalist, with some higher tones forward in places and supported notes, and have been noticing on certain range, a spot once in a while, where it will make that slight fuzz, barely detectable, only with headphones. Couldn't master it out, had to re-record that line. I'm thinking to graduate to a new USB condensor mic with 24/96, having several songs to record the vocals on and on sale. ..and hoping it will give me more "presence" . (I have several gadgets-apps to add width etc....) Appreciate your video.👍 This helped clarify the difference... I'm still undecided to spend the money now, which I will if I have to...👍ps: If you have any extra input for me on the subject...much appreciated!
I do a tremendous amount of vintage audio restoration. The default for me is two channel 24/96 because of the amount of manipulation of the audio to make it pristine. My final product is 16/48
It's all in your head ;) 24 bits resolution (bits/sample) is better. Yes. Period. 96KHz is just stupid if you know the mathematics behind analog-digital-analog conversion. Because we can't build perfect output filters, using a higher sample rate than the theoretical sample rate will resolve eventual aliasing noise in the higher frequencies above 21KHz and close to 22KHz (which 99.9% of humans can't even hear)... But 48KHz is more than enough and with very good output filters you can use 44.1KHz perfectly fine for human hearing.
Hey there! I discovered your channel while searching about audio bit depth, and I am loving your videos on this topic. Also because I a studing electronics Engineering, and I am learning about Sampling Theoreme, pretty cool coincidence. My comment is to support your channel and also to ask you about sources of information that mix this topics with music, to go a little bit further. Thanks!
i bet it just must be upsampled on stretch algo level. or even better - we should have ability to select oversampling for stretching anyway great vid, thanks!
Technically, if you look at the behaviour of the tweeter of your speaker, and if it is capable of moving above 22kHz, it will make a difference in clarity of the high frequency content played. similar like the front plate of a guitar body (where the strings are attached), you get some kind of Lissajous-Figures. And hence the audio will be distorted - whether that's audible or not, that's another question.
Thank you so much, such a highly informative video. I think one point you could have touched upon would be that if a file is going to be slowed or pitched down, that a higher Sample rate file might also sound better because a lot of the before inaudible stuff moves into the audible range in contrast to a 44.1khz where it capped at ~20khz and after lowering pitch or time, the cap also moved lower creating a high cut and leaving the high shelf empty making it sound dull.
I agree that higher rates sound better when editing, but the original, auto-tuned vocal recording was so bad, I mean, who cares about the stretching resolution? GIGO.
Which audio interface you are using? I recorded at 48k and have found recording at 96k definitely improved my sound. I don't do any time stretching. At 96k I had better stereo image, cleaner higher frequency and better low end.
Remember ufx and a minidsp flex connected by adat. Very strange! Is the difference in raw tracks or processed tracks with plugins? Have you tried down sampling the session to 44 khz and then doing a blind shootout? I've never been able to hear any statistically significant differences...
@@RaytownProductionsThank you for your reply, the difference I heard was even on the raw tracks without any processing plugins. While it's true that humans can't hear anything above 20 kHz but they can hear the audible difference when capturing at higher sample rates. It is to do with the DAC and how good your converter is. At lower frequency the converter tends to produce unwanted distortion which will be avoided or minimised at higher frequency. Don't take my word for it please read what Bob Kats wrote on this subject. Create a session at 48 kHz and just simply play any instrument low to high octaves through your audio interface and then change it to 96k and tell me you can't hear a difference.
The primary advantages of recording and/or editing at rates above 44.1kHz are: 1. The ability to properly filter frequencies above 20kHz without losing the audible frequencies in the 15k-20k range. This makes it much simpler to eliminate aliasing. 48kHz or high is usually sufficient for this if you have analog filters that cut most content above 20kHz in place. Going to 88.2k or 96k for recording makes it much easier to filter effectively. 2. You can do more accurate frequency adjustment. (Time stretching) 3. You have a marginally lower noise floor, although higher bit depth is much more effective at lowering noise floor (while also adding dynamic range)
Great points. I agree with everything you say here. Digital filters are REALLY good now so in my opinion it would be really hard to hear the differences between an antialias filter at 44 and 48. Thanks for the comment!
@thinkingfield good point! But it doesn't always work that way. The trick is that the aliasing needs to be removed BEFORE it is folded back into our hearing range. This is why it's important to do this within the plugins right after the audio is internally upsampled.
Good stuff, Bobby! On another note. Which one of your previous videos shows you using the Lindell Series 50 API plugin? I dug through your videos, but couldn’t find it. Thanks! One of my favorite channels by the way! Keep it up!
Well 2inch reel to reel tape with the best Dolby -s reduction was around 13 bits ! so 16 would be better, and so 24 bit as well and 44.1 or 48khz, so really if you want to sound better be like van halen if you can, and you will sound better !
Honestly… I would find the high sampling rates to be a complete gimmick, if it wasn’t for the time stretching. They seem to create more issues, than what they are supposed to solve. I have a MOTU M4 audio interface and even though it is my best interface in terms of signal qualities, running it at more than 48kHz extends my noise floor into ultrasonics. Try to clean that up. It will make you go nuts! By the way, the sampling rate only defines the highest possible sampling rate. The noise floor / dynamic range is defined by the bit depth, depending on the recording format. PCM WAV for example: 8 bits = 48.16dB, 16 bits = 96.33dB, 24 bits = 144dB, 32 bit = 192.66dB.
Oops! I wrote sampling rate twice there. I obviously meant, sampling rate defines the highest possible frequency to record, before aliasing occurs. And writing about the dynamic range: a noise floor of -48.16 dBFS is almost impossible to clean up. You need to destroy your audio with compression, to fit it in there not even the human voice does fit in there. Funny enough, a noise floor of -96.33 dBFS again is so low, that it makes for great releases. But for recording, having the extra low noise floor is great for dynamic range, even for the human voice. I can easily record whispering and shouting in one go. That being said, 32 bit PCM remains a gimmick, because for now, there is pretty much no audio recording equipment on the market, that has a dynamic range to surpass even the limit of a 24 bit noise floor. 32 bit float interfaces again are realized with trickery.
nice! i'm definitely going to upsample when i edit now, that seems to solve the problem of "oops i picked the wrong algorithm this sounds bad", especially when going from electric DI to bass or acoustic where it matters a lot more
Did you went to a recording school? I was wondering cause if you do. You would know that 44,1khz and 96khz are incompatible. 44.1khz and 48khz are two different type. To record higher sample rate for 44.1khz you would use x2 to 88.2khz.
I hear a very subtle difference in the very highs using my objective calibrated ears according to a scientific reference. Assuming there is a subtle perceivable difference, I think at least the following factors contribute to this 1. DAC quality/flaws 2. DSP algorithm flaws 3. Audio format That excludes engineering/production quality, hardware, opinion, differences in hearing perception, audio file source quality, streaming service and settings, computer/phone/amp, build quality, speaker cables, cross-over quality, and driver and cabinet quality. I think in theory 44.1khz is sufficient, however in reality a higher sample rate may objectively sound better, although it's extremely subtle. It's possible some DSP algorithms, whether it's in your amp, computer or plugins work better with higher sample rates. Perhaps inversely for others. Therefore perhaps the average person's audio setup would benefit subtly from an increase from 44.1khz. While a perfect audio setup may benefit less. I think we're not as aware of the layers involved in modern listening. It has never been this complicated with so many variables that are also abstracted away. Sure one may consider upping the quality of their streaming service, however, how many consider how their chosen streaming service compresses and delivers the audio? That's not inherently a bad thing, but more awareness could help. Perhaps that's why some people who try a well-recorded CD say it somehow sounds better - because we mostly listen through advanced digital devices with many layers, while a good CD player connected to a good amp has far less.
I make part of my living recording overdubs first different recording studios from my remote studio. Most of them ask for me to record at 48. But I have one client that always asks for 96. I do what they ask. But when I do my own stuff I usually just record at 48.
I'm glad I came across your channel whilst 'researching' headphone mixing plugins because this is another great video! Clear explanation and nice to see it summed up in one video. Might I ask what your thoughts are when it comes to bit depth? I know 24 bits gives a larger dynamic range so I can understand why you should/could prefer it over 16 bits but what is the point of, when I look into Cubase 12, 32 bits and 32 bits float and 64 bits float? 24-bits already gives a dynamic range way beyond what we can hear it seems :) Curious to what your opinion is (or perhaps make a video about that too ;)) Update: Ha ha, I just looked at your video list and already see a video which seems to address this. I'm watching it now ;)
in the samplerate conversion phase, ITB, it doesn't matter the adc/dac quality but the audio engine of the DAW, for example the rendering in Ableton is the worst one, so I we could test the quality of SRC in our DAW ? Or must we use, outside of Daw, sowftare as RX of Izotope to this specifica task ? for one track ok but when we have many ones isn't ok to work smoothly if we must resample (two times too) outside the DAW. Is there any high quality src plugin vst ?
There is one place where this matters more than music and that is in sampling and sample stretching. The higher the sample rate, the better the sound will stretch in a sampler, for example, Omnisphere (when you want to import a sound and then play it across the keyboard).
That is just one reason why it bugs me that most sound design companies sell samples at 44.1kHz. I mean that WAS the audio standard in the 2000s but we're not cutting CDs anymore and stretching across the keyboard definitely benefits from higher sample rates. :)
@@JustinLesamiz Um, what? So you believe that those of us who use samplers use them because we don't care about sound quality? What do you suggest then when we are all limited by the technology and proliferation available? Mmmk. Good luck with that line of reasoning.
Because 48kHz moves the required high cut in the audio signal from 20,5 kHz to 24 kHz, which is in reach to the frequencies some people can hear, while 96kHz moves it from 24 kHz to 48 kHz, which is irrelevant for human hearing.
Have you tried a blind test? It's very easy to fool yourself into hearing things that might not actually be there. Try a null test first. You can upsample the 44khz song to match the 96khz song and invert the polarity. IF it's silent, there is absolutely NO difference. Use a good resampling algo like r8brain by voxengo (free). Then do the same test by downsampling the 96 khz to match the 44 and invert the polarity. Do they null? If so, they are IDENTICAL and the differences you hear are in your mind. Also, if there is some sound during the nulltest, it needs to be pretty loud to actually hear it when the song is playing. Let me know how it goes! Thanks for the comment :)
Now when you say editing - what if you are just comping a vocal? No time stretching. Just choping and moving sections forward or back to get a perfect vocal but not stretching and doing minimal effects? Is it still recommended to record in 96 or upsample to 96 for editing?
There is a moment in your video where you say “I don’t know”concerning why should you record hypersonic frequency because no one can hear them. In fact, this was done by a research team led by Manabu Honda in Japan and published in the Journal of Neurophysiology in July 2000 under the title "Inaudible High-Frequency Sounds Affect Brain Activity: Hypersonic Effect". There he proves that, even inaudible frequencies will be processed by our brains. The enormous brain power that comes about is particularly evident when it comes to the localization of a sound event. A sound coming from the right side of the head is picked up by the right ear first. The sound travels further around the head to the left ear, only to arrive there a little later and with a slightly different timbre. The brain uses this minimal time and sound shift to recognize the direction of the incoming sound. The temporal resolution of our brain is in the range of 5 to 10 microseconds = 0.000,005 seconds. So it’s not only about frequency but also about timing, therefore 24-bit/96 kHz are needed if you want to max out quality for those with good hearing. However, I agree that highly compressed Pop music won’t greatly benefit from it, music from a classical grand Orchestra would.
Thanks for the reference. I'll check out the paper. And just be cautious with scientific literature like this - the scientific community presents evidence to support a hypothesis. It doesn't ever 100% prove anything empirically. Our understanding evolves as we learn more and more. I'm actually guilty of this by not saying "the science suggests" instead of using absolutes like in this video. Appreciate the comment! Cheers.
@@RaytownProductions I know you have the point of view “recording“, but there is also point of view “listening”. A DAC uses digital reconstruction filters to create the analog sound. Those filters are problematic and create unwanted artifacts, like pre-ringing, Aliasing-Errors and other nasties. With Hi-Res some of the problems are shifted into the hypersonic area where they do less harm than with 44.1 kHz. That problem is even better solved with DSD than PCM.
@@goodsound4756 Good points! Pre-ringing would occur with linear phase or mixed phase filters. Of course you will get some phase distortion, but humans don't seem to be sensitive to it at the low levels we would expect from a reconstruction filter. Also, anti-aliasing filters are very very good nowadays leaving almost no artifacts. Check it out here: src.infinitewave.ca/ I've done LOTS of blind shootouts trying to see how much aliasing can occur before I can statistically pick it out in music. Any aliasing that occurs from something like that is so insignificant - it needs to be about 100X louder for me to reliably pick it out. Could it somehow change the perception of music? Perhaps. But I'm skeptical until I see more data. The paper that was mentioned seems to be reasonable, but I'm not a neuroscientist so I can't say for sure if the methods are sound (no pun intended). The number of data points are also pretty low and there aren't any concrete conclusions - more so just "our data looks like this, but we don't know why". Interesting conversation nonetheless. Thanks for the comment :)
32 bit 96khz should be the standard imo....your ears are influeinced with super tweeters but they don't even start till 14 - 16khz...we need the bandwidth...& for bass
Depends on if the analog outboard gear can handle it well. A lot of the downstream gear is just producing more distortion when fed ultrasonic content and that's the 'pleasing' distortion you might hear.
From the mathematics and physics on the subject instead of an Audio Engineering perspective, the math and science community actually model the theories and then measure the results to confirm or disprove theory. Let's start with the Nyquist. The OP mentioned filtering out the frequesncies above 1/2 of the sample frequency fixes the Ailiasing once and for all. This is actually not true. When mixing two signals, in a non linear mixer, the two frequencies and the sum and difference frequencies ARE produced. There is no magic that happens if the one signal is above or below the other frequency. The higher the bit depth, the more linear the result is, so the Ailiasing becomes minimal, but still exists. The Nyquist frequency for CD quality is one half of 44.1KHZ or 22.5 KHZ. Bit depth, believe it or not is DYNAMIC and changes in real world audio recording. Here is how that works. CD quality is 15 bit. Record a signal 6 DB below 0DBFS and your Mist Significant bit is not changing, so effectively you now have a 15 bit recording. Drop another 6 DB and you drop another active bit for a 14 bit recording, etc. In the math world and audio world, a 3 DB drop is a drop to 1/2 the energy. a drop of 6 DB is 1/2 the VOLTAGE, or the value of one bit in the A/D process. The reason mastering is not done professionally in CD quality is any headroom in the recording results in the same amount of loss of active bits in the 16 bit A/D conversion. With a 24 bit depth master recording, you can preserve that 12 db of recommended headroom and still have 21 of 24 active bits when you normalize the recording up to 0DBFS. Remember to have 12db of headroom, you leave 1 bit inactive for -6db, and 2 bits for -12 db, and 3 bits for -17db. With the higher bit depth, the amplitude of the Ailising becomes minimal. To eliminate it in the master recording session, it is as simple as doubling the sample rate. A 86KHZ recording at 24 bit depth will provide the following. The Nyquist frequency is now 48KHZ, Remember the SUM Difference and 2 original frequencies? The difference frequency of a 20KHZ sound and the 48KHZ nyquist frequency is well above 20KHZ and very easily removed as it was never in the audio range below 20KHZ. The amount of non-linearity due to the 24 bit depth is very low, so it becomes a non issue for the original master recordings. It's in the math, it's in the physics, and it is in the sound. For affordable equipment, and hood low to midrange bands, an excellent compromise is to master at 48KHZ 24 bit if you are not mastering for a CD. You don't want to add difference frequencies in your master by adding the difference of 48KHZ and 44.1 KHZ into your CD mastering. This is why most budget minded digital consoles are switchable for 44.1 and 48 KHZ. Live sound only, use 48 KHZ. Recording for streaming, RU-vid, etc, use 48KHZ, Recording for a CD release, master in either 96KHZ or 44.1KHZ. There are many budget friendly 32 channel digital mixers with 24 bit depth available for use with a 32 channel DAW and digital snake to the stage. The myth of metering mentioned by the video is a joke. When is the last time you saw an audio meter with 24 or more segments to represent the full sampled waveform. Most of the LSB bits are scrubbed from the meter bridge and the MSB bits are represented and a trigger on the highest bits hold the Peak value. If the peak is not caught by the recording, it is not in the recording. A repeating cycle of music will have many samples to capture the peak value of the peak reading meter. Be sure the Peak that comes on is never the CLIP LED. that comes on on most consoles when the most significant bit is reached indicating you are in or exceeded the -6 db bit of headroom. To get an inter=sample peak of any significance, the frequency in the 20 KHZ range would have to be exceedingly high.energy the energy in this frequency range is minimal. The discussion on the subject is limited to the math scholars and not anyone running a session with any proper headroom.
Appreciate the super thoughtful comment... You make some great points but at the end of the day, can you hear a difference? Especially given that most plugins now offer oversampling. I may be a bit skeptical, but that's ok! 🙂
Great video. I just have 1 clarification if you don't mind. If a 44.1kHz file can be upsampled to 96kHz and be treated like it was 96k in the first place, how is it different if I just record at 44.1kHz and just upsample later for editing (assuming I don't really need to record frequencies greater than 22kHz)? Since ISP and time stretching artifacts can be handled by upsampling, I am wondering if it even matters that something is recorded at just 44.1kHz.
This is a good question. From what I can tell from my experiments and testing I did, you can probably just record at 44 or 48 kilohertz and up sample just for editing. Then I recommend down sampling back to one of those formats to do the mixing in mastering. If your plugins offer oversampling you then have the best of both worlds.
@@RaytownProductions Cool! Thanks for the reply. To be clear, I am not refuting your advice though; it is still good advice to go 96 if you can, but if you forgot, it is not the end of the world. And as you said, there are plenty other things that matter to a good mix than the sample rate. 😅
@@antonm_ totally agree. The higher rates can definitely help with aliasing reduction and lower noise floor (it's complicated how that works, but it's really something that happens), so I totally agree. If you can swing it, 96 can sometimes be beneficial. 🤘
Ciao, great work ! A question, wich kind of edits, apart time strecthing, is rilevant to prefer higher sample rate ? If have to upsample to 96 isn't better to work at 48 than 44.1 (2x multiplier..) ? If is better to upsample why do not record at 96 from the beguinning ? Because for tracking we get lower latency too; ok cpu work more but today is not a real problem, thinking to record an instrument with only a vst (rev or channel strip) on performing over only a single stereo audio file (temp mix as "karaoke"..) with no other vst on. So at the end the other option is oversampling that can give preringing issues with equalizers, solved adding a Q and small bump at the cutoff, or I a missing anything ?
Thank you for this. I'm on the fence between 2 digital mixers: Yamaha DM3S which supports 96kHz or Mackie DLZ Creator which has 24bit/48kHz. My main use would be home music studio work. I write music and play every instrument. I feel like the Mackie would meet my needs even though it has less inputs, it's $900 cheaper than the Yamaha. Do I need the 96kHz to write an album or would I be ok with 48kHz?
Wrong wrong wrong. 96k samples is more accurate. Take a session and do 44.1 then 96 and there is a huge difference in how the plugins react and accuracy. The music sounds wayyy better when i work at 96.
Strange. This is not my experience at all. What DAW are you using? What plugins? Did you do a blind shootout between the 44 and 96khz renders to confirm that the 96 khz session sounds better?
@@ankeviousoliver4472 interesting! What were the plugins you were using? Sometimes if they are poorly coded or don't offer oversampling, the higher sample rates can help reduce aliasing IF they are a non-linear process.
@@RaytownProductions i mostly run waves plugins and dont use much saturation or oversampling for the most part. I do run some izotope as well. One thing i noticed is that some waves plugins cant handle 96K though which is my only issue.
Wouldn’t it be better to use over sampling than to use higher sample rates? In my case I use more non-linear plug-ins than I use time stretch, and inter-modulation is worst with higher sample rates (unless you filter anytime you use a nonlinear plug-in).
Great vid. Explains everything. Is there a system or plugin to correct the metering problem. The sound is generated by the daw so why not just do the up-sampling when needed if a peak is high enough to be inter and calculate the true peak from there for the meter?
Soft synths sound better at 96 k for whatever reason ..i can A/B it now and hear it plain as day ..I use tempos derived from whole milliseconds ...so 48 per millescoend works for me ...easy edits with the right clock ...96 for the soft synths though. what about clusters of small frequencies high up ....it affects the pressure wave. I think Neve said his kit was capable of 100k ..and we can't hear that far up ..unless in a hypersensitive state. I prefer analog for that reason. I mix at 48 though. It's a timing thing ..not a perception thing.
Your own diagram is almost self explanatory though, in your first diagram you pretty much illustrate the meaning of a quantization error in the data recorded, if you were recording a square wave at 22.05khz there’s a 1 in 4 chances that the sample is out of phase when played back at 44.1khz, and there’s a really high chance the audio is distorting because the peak isn’t being capured. The higher the sampling rate, the more accurate it is because the chances of true peaks being actually captured increase, and the quantization error distortions become a lot less pronounced. On the first figure you show you can pretty much figure out visually how one of the sampling points is not being taken from the apex and from there the following point is captured under it, when it gets played back, the amplitude of the signal will peak that much lower, and from there not only will it not follow up with the intended tone but it will drop to a lower value. The worst part of this is not that effect per se but the fact that it’s random, the sample rate clock doesn’t sync up with the timing at which you hit record, the only way to solve this, is by recording at a much higher sample rate.
I asked chat GPT about this, this is the answer, "Is there potential audio degradation from recording something a sound near 22.05khz at a 44.1 kHz sampling rate?" ChatGPT "Yes, there is potential audio degradation when recording something with a sound near 22.05kHz at a 44.1kHz sampling rate. This is because 22.05kHz is half of the 44.1kHz sampling rate, which means it is exactly at the Nyquist frequency. According to the Nyquist-Shannon sampling theorem, the sampling rate must be at least twice the highest frequency in the signal to avoid aliasing. Therefore, in this case, the highest frequency that can be accurately sampled without aliasing is 22.05kHz. Any frequency above 22.05kHz will be aliased to a lower frequency, resulting in distortion. To avoid this issue, it is recommended to use a higher sampling rate, such as 88.2kHz or 96kHz, if you want to capture frequencies near 22.05kHz without aliasing distortion. This will allow for accurate sampling of frequencies up to 44.1kHz or 48kHz, respectively, without aliasing." It is recommended because that's the only way to actually get rid of most errors (it really never actually gets rid of the error for as tiny as the may be, except they tend to blur out enough to be imperceptible, however, as there could be a lucky shot where the clock syncs up perfectly with the input and you get almost 0 distortion while still recording at 44.1khz, it's still a probabilistic concern you can avoid thinking too much about at 88.2 or 2x 48khz respectively as it said.
Unfortunately chat gpt almost got it right. You must sample at GREATER THAN half Nyquist. Be careful using generative ai. They can "hallucinate". That with confirmation bias from the user is a very dangerous combination. That being said, most of what you're saying is correct. Lots of weird stuff does in fact happen near Nyquist. The example that you gave with a square wave is one of the most extreme examples and not really applicable to most music, not to mention that square waves don't exist in reality - they require the summation of an infinite number of frequencies.
Logically I understand down sampling, but how can a already sampled recording be up sampled? Is the existing (already recorded) 44.1 sampled sound reprocessed? Isn't that logically less quality if your resampling a sound that is already a lower resolution, compared to a higher sample of a raw, unsampled recorded sound?.... I am not criticizing, I am just honestly asking.
Typically the missing samples are "zero filled". This topic gets complicated quickly and I'm not a digital signal processing expert. From what I understand, the zero filling just extends the frequency range that the audio processing can work, among some other minor things like improve amplitude resolution. This will cause "mirrors" of the audio spectrum at higher octaves which are filtered off with a low pass filter called a reconstruction filter when down sampling back to native sample rates like 44. This up sampling won't really add back any new frequencies, it just allows a bigger frequency window for processing non linear plugins which helps to minimize any aliasing that might occur. This low pass filter is important to do, otherwise you will get lots of aliasing if you leave that higher frequency component in the audio. This is usually handled by your sample rate converter so you never have to think or worry about it. Hope that helps (any DSP engineers feel free to correct me here haha!).
Does all of this matter for vocals as well? I’m a rapper and I am looking to get the best quality out of my music when I mix & master my work. What is the best sample rate to record for an artist?
Hi, you said that there might just be one possible solution to have a signal between 2 measuring or sampling points. That's hardly believable. If you have a 9kHz signal you will have 5 sampling points per period. But what is if there is an additional 18 or 20 kHz signal of very short duration. it would change this signal and the resulting signal might not be captured accurately, I guess... Hard to explain in a foreign language And: do you think it is possible to convert a 48kHz audio into 44.1 kHz audio ( or vica versa) with out a loss of quality? Or is it necessary to record in a double- target sampling rate? So 88.2 kHz for CD instead of 96kHz? Kind regards
Digital signal processing isn't an easy thing to grasp. Assuming you sample at greater than 2x the highest frequency and the signal is band limited, it will ALWAYS be perfectly captured. Period. It's mathematically proven. If you want to go deeper, you can read some digital signal processing books that will probably have the mathematical proof. It's not intuitive, and it can take a lot of reading to finally get these concepts to click. It did for me at least 😂 Assuming a high quality resampling algorithm (which almost all DAWs have), the only noticeable difference will be the loss of frequencies above the anti aliasing filter (which is about 22khz which we can't hear). There really isn't any benefit to recording at such high sampling rates unless you want to reduce aliasing in non linear processes and plugins. If the plugins have oversampling, then it's honestly probably not worth the extra cpu unless the plug-in oversampling algorithm isn't working as expected (which can happen... I've seen it on more than 1 plugin!). Hope that helps! Cheers 🤘
Please ask, sir, this is related to the Type C to Jack Audio Converter Adapter which already has a DAC, there are several options on the marketplace, some are 24bit/96 khz, then some are 32 bit/384khz, which one is the best at capturing small sounds but sounds clear which one bro? this is for the case of playing PUBG games to be more sensitive to listening to distant enemy steps.
In that case you would want bit depth. The only real advantage I think you will have is going from 16 bit to 24 bit. 32 bit would in theory be the best but I'm not sure you have your speakers loud enough for this to make a difference. It's very rare that you would perceive an increase in quality from 24 to 32 bit.
I heard somewhere that it is better to create projects at higher rates because if TV/film etc want to use your music they demand higher rate. Is this the case?
I tend to go deeper than other channels out there so you can fully understand the concepts instead of just getting a yes or no answer. You will find most of my videos run on the longer side, but the concepts are fully covered so you don't need to watch 4 or 5 videos to get a full answer. That being said it's always good to hear feedback and I'll do my best to give complete answers as quickly and clearly as possible.
I’m fact using focusrite interfaces setting anything besides 48kHz caused issues for me I would get static and no audio at times set it to 48 no issues
Aw damn this just happened to me! Seems this was why. Except I’ve mostly recorded at 44.1, no issues. I think there’s some setting that’s fixed on the I/o software - mine is stuck at 44 pretty sure. Could be related
Good point, but why Audio Interfaces Offers lower latency in 192khz? If u look at the latency benchmark all brands ask you to increase the sample rate and decrease the buffer size. For an example Motu M4 offers 2.5ms roundtrip latency at 64buffer size / 192khz. If you go 64buffersize / 48.000khz the roundtrip latency ups to 4.3ms. Can you talk about it?
Most modern ADC use two stage filtering when running at 44.1 kHz or 48 kHz sample rates. When you select 48 kHz, your ADC will operate at 192 kHz internally, which allows for gentle analog Nyquist filtering. This captured 192 kHz data is then run through a steep digital linear phase filter before conversion to the desired 48 kHz output. Importantly, all linear phase filters need latency to operate. However, if you select 192 kHz, the digital linear phase filter stage is disabled, thereby removing the additional latency requirement. Of course, your ADC miles may vary.
@@thomdabomb5067 interesting. I'm a type of producer that use VSTs in 90% of my production via USB midi Keyboard, I record only vocals and sometimes a guitar or single instrument via audio channel. Also I am those ones that mix my songs on the flow (while produce). So, having these aspects in mind, which sample rate / bit depth/ buffer size is the best for me?
@@laserjakk3629 Bit depth is easy: 24-bit for studio recording. (If you're doing field work, those fancy-schmancy 32-bit floating point recordings like the Zoom F6 are amazing!) Buffer size will depend entirely upon your setup. You'll just have to experiment and see how small you can go while still maintaining a margin of safety. Sample rate for recording (where ultrasonic frequencies are not needed) is an interesting topic. 44.1 kHz is fine for things like music streaming, but 48 kHz is used for video (although RU-vid notably recommends 44.1 kHz.) However, you'll read/hear many people claiming that 96 kHz or 192 kHz recordings sound better, especially in the high end. I'm assuming you're aware that audio reproduced through digital means is not jagged, and that sample rate is not about "resolution" but about frequency response. Does this mean that claims of 96 kHz or 192 kHz recordings sounding better is just a placebo effect? Maybe. But maybe not. As previously stated, if you're recording at 48 kHz, there's a good chance that your audio interface is running internally at 4x that rate, and then converting down to the target rate of 48 kHz. But there's no one correct way to do this, and some interfaces do it better than others. If you find that your interface captures better at 192 kHz or 176.4 kHz, then by all means use that. However, after capture, I would recommend converting those files to the more useful rates of 48 kHz or 44.1 kHz. But be aware that like audio interfaces, not all software sample rate conversion algorithms are equal. I personally recommend something that uses the latest r8brain algorithm by Aleksey Vaneev of Voxengo fame. He uses it in his plugins, and last year Reaper added it as option. (Mr. Vaneev has kindly made the source code available for all to use.) If you're not a Reaper user, you can still download its fully functional and perpetual demo. Under the "File" drop-down menu there is a "Batch File/Item Converter" option that allows you to do all sorts of conversions, including sample rate conversion. Be sure to select "r8brian free" for optimum quality!