How Sample Rate Conversion Effects Sound Quality. 

Great test! It is important, though, to mention, that distortion, artifacts and noise, even at extremely low volumes, can have an influence on the material at normal levels. It is similar to applying dither noise when reducing the bit depth. That noise is near inaudible, down at - 90 db's (16 bit) or lower (in case of dithering to 24 bits), yet it is well documented, that changing the dither noise by noise shaping has an influence in the sound. Yet if you do the phase inverse test of the two files, or also dithered vs. non dithered, you have to crank up the volume even more than 48 db's to really hear the differences in noise. That said, Izotope does a really great job in the domain of SR Conversion.

Very useful, thanks! I'd like to add that Apple has one of the better sample rate conversion algorithms, inducing very little aliasing. I guess the result would probably be much noisier using 16 bit files with dithering, but your test is still valid and we should stop worrying about sample rate conversion. The only time when you should care about sample rate (in my experience) is either if you want to stretch the recording without losing high-end details, or using distortion/saturation plugins that can cause some ugly aliasing if they don't have oversampling. And even in this case, it remains pretty subtle.

I know theoretically a null test shouldn't produce any noise, but would it be prudent to do a null test on just the original audio and it's polarity reversed version to make sure the null test itself doesn't produce any noise?

Thanks for the video! I'snt Cubase updated and trasnpartent now?

Very nice! I expected 400x to sound much worse... I've only had patience to do 10x and couldn't hear a difference. Modern sample rate conversion is awesome! There are few more things to consider thou: if you use a good modern converter program, it doesn't matter in offline conversions if you convert from 44.1 to 48 or to 88.2 or to whatever, they all sound good (as the video proves in practice). And they also all go "even". But it takes more CPU to process "uneven" sample rates because the Least Common Multiple (LCM) is a much bigger number. When the target sample rate is half or double, the number is small, for example from 48 to 96, the LCM is 96 and it just needs to be multiplied with 2. But when converting e.g. from 44.1 to 48, the algorithm looks the LCM which is 7056 meaning first the 44.1 is multiplied by 160 to 7.056 mHz, then processed and finally divided by 147 to reach 48 kHz. In offline conversions the only downside is it just takes a little bit longer but in realtime conversions this may be a small problem. Take for example Kontakt. The choices are either to drain the CPU or another, worse but lighter algorithm for "uneven" conversions (44.1 -> 48 for example). Kontakt has chosen the worse algorithm which is understandable. This can be noticeable especially with drum libraries, on cymbals.

A question that rages week in week out in the world of video is that recording 44.1KHz causes video drift when the end system is 48KHz. I've done my own tests and not found this to be the case. Clearly if they were interpreted incorrectly that would be 8% and would mean a speed change. Final Cut Pro defaults to 48KHz timelines and any content is conformed to this on the fly so 44.1Khz is converted to 48KHz. I think it does a pretty good job of this. My conclusion is that audio drift is caused by the accuracy of the respective clocks in each device, especially if they are unlocked and free running. Anecdoetely maybe it could be said that 44.1KHz kit tends to be cheaper and maybe the crystals in the clock keep worse time? So sample rate conversion as your test shows here is pretty seamless. Avid Media Composer used to have 2 modes 1) Fast and Poor 2) Slow and Quality, neither ever changed the speed in my experience. There are some interesting facts around whole samples per frame: 48KHz/30 = 1600 44.1KHz/30 = 1470 48KHz/25 = 1920 44.1KHz/25 = 1764 48KHz/24 = 2000 44.1KHz/24 = 1837.5 so it's only the 24fps 44.1.KHz combination, that are not whole (if you discount 23.98 and 29.97 which clearly aren't). Could it be a factor when locking frames to video? Audio doesn't have a frame rate but when it's recorded along with video it is locked to it, does this matter when trying to keep sync and /or sample rate converting down the line?

I’m interested in the different methods of sample rate conversion. Can you show how you actually converted the samples? Was this in logic or with a third party SRC?😊

Very good myth-busting.

Sorry but who cares about the noise floor, not exactly an issue 2023? Maybe I'm misunderstanding something here what you're trying to achieve but higher sample rate is only going to affect the higher frequencies. And sure, that will be perceived as a lower sound quality if put to the extreme. So basically I would be just perfectly fine with recording let's say a kick drum in 22 khz.

All the audiophiles "experts" screaming in the comments about the difference 😂 Where are they

2:50 I can hear the difference, the second track has something like hifreq eq boost or transient artifacts on it and the chord at the end has some artificial wobbling and quality loss or artifacts. First string strike transient is also louder in second track.