When doing sounddesign for games & movies, sometimes you wanna do extreme pitch-downs.
If the source sound was recorded (sampled) with lower rates like 44.1 or 48 Khz, then the Nyquist frequency becomes audible; you’ll hear an annoying tone through the pitched down sound.

Therefore when doing extreme pitch-downs, it's best to use sources with higher sample rates of 96 or even 192 Khz. With these rates, the Nyquist frequency will stay out of the audible range. High sample rates also work better on extreme time-stretching, which was already mentioned.

Read also:
https://en.wikipedia.org/wiki/44,100_Hz and
https://en.wikipedia.org/wiki/Pulse-code_modulation

 

Makes sense...

 

48 / 24 if you creating for video
doesnt matter otherwise......fact

 

Because it is much more than enough, it is for safety, either for subtle or hard distorting. For example, i don't hear so much aliasing at 44.1 with or without oversampling, because there is a lot of IMD, odd harmonics etc. If you have a lot of highs, a lot insane boosts at highs, than add a lot of distortion - yes, you will hear that no oversampling is noisier than 2x oversampling. 2x vs 4x and even 8x is way subtler, almost i don't hear too much difference. Because aliasing is messed up with intermodulation distortions. So, for 96k the oversampling will be subtler, way subtler.
Synths like serum do a lot of difference at highly pitched notes with vs without oversampling. Sine waves too. But when you use distortions or saturators, it is way difficult to differentiate aliasing vs intermodulation stuff. But if you hear a lot of aliasing and your oversampling solves problems, means you do something insane and physically impossible. For eqs and filters, while working at highs you should hear the difference of cramped/decramped filters, oversampling on vs off, equing highs at 44.1 vs 48 vs 96.
So even for me, using all this shit is enough or subtle or no difference with 96k and 2x oversampling.

 

Plugindoctor
Set sample rate
Choose oversampling.
Drive all the shit out of it. And you will see.
Usually you don't need oversample too much with high rates, but oversampling in most plugins works fine. High rates and 64x oversampling or even 512x give excellent antialiasing results. But do you need them to be reduced if they are below -90dB? If you have a lot of aliasing and they are too loud, you do something insane or physically impossible.

 

There is a very nice plugin ExperimentalScene ES AntiAIias. It is free. It is a plugin that filters out harmonics that occur above the Nyquist frequency by oversampling the input signal and then filtering. Has 64 Bit Floating Point Audio Engine and Minimum Phase Distortion FIR Filter. The oversampling range is from x2 - x32. By changing oversampling rate of this plugin while processing the same file using the bypass button you can easily see that one hardly can say that there is some "sweet spot" under x32. Of course, that is just an "ear test". I don't know what Plugindoctor would say, I don't have it installed yet. Also I have heard examples, when on some files only with x64 you can hear delicate vibrations within reverberation.

 

Depends on if you can hear some characteristic artifacts edginess and blur [especially in high end] and to what degree. Usually, you can do with just x2, for the sake of security without much thought. A general rule is the higher sampling rate, less oversampling is needed, but it is just a loose guideline.

 

Last edited: Dec 3, 2019

I also like this one:

 

I can confirm this. For years I worked with 44.1\16bit, then tried higher SR with no difference, then tried 44.1\24 and the difference was stunning...more detail and clarity in the entire audio spectrum. Higher SR with 24bit had no >audible< difference from 44.1. Only problem is having to transfer things to DVD standard which is 48\16.... upsampling is far less error prone than downsampling, and a good dither after 24 to 16 bit conversion keeps things in excellent condition.

 

I wonder how this will be possible.

 

I'm surprised at statements like that. Here amplitude and dynamics is compared with frequency modulation, as if you could fight for an absolute truth.
Playing the winning properties of bit depth and sample rate against each other is like comparing equalizing to pitchshifting. One is no better than the other. They serve a different purpose.

For bit depth, let your converter sinc the material at 24bit. This increases the precision when sampling the amplitude and at the same time increases the maximum dynamic range by 50% compared to 16 bits. Of course, a higher dynamic range also gives you advantages when processing with dynamic tools such as EQ and compression.

The sample rate is similar. But to explain the process of digitalization is much more complex. The simplified explanation is that at a 44.1k SR 44100 sinc poits per second are created. Based on the temporal and dynamic position of the sincs, the analog waveform can be replicated by further interpolations. At 88.2k the sample rate and the number of subsequent interpolations doubles, which also increases the cpu load. In short, the bandwidth of available information doubles.

This always plays a role in the further process when frequencies are modulated or generated. A clear difference between 44.1k and 88.2k can be heard not only in distortion, but also in all forms of pitch and frequency modulation. The more drastically this processing is performed or the closer this processing takes place to the limits of the available frequency bandwidth, the more obvious is the difference between the sample rates.
This means that the project determines which sample rate is most suitable to you.
While a sample rate of 44.1k can be sufficient when using just clean EQ and compression, a sample rate of 176.4k or 352.8k is the wiser choice for excessive sound design.

A subsequent upsampling only doubles the cycles of interpolations, but the required sinc points are not available to guarantee a corresponding precision. It's just a method where some artifacts are accepted to reduce other artifacts. It's a compromise.

 

Last edited: Dec 3, 2019

The guy was not talking about technical comparisons, but about the audible impact of different SRs/bit-depths on the overall sound quality.
And he is absolutely right. It is what you hear that matters in the end. The truth of aural perception does not depend on the art of engineering analysis.

 

It wasn't just directed at him, but I'm too lazy to search the whole thread for similar statements.
My point is, depending on the overtone bandwidth of the instrument and frequency range of microphones and amplifiers in front of the converter, you can definitely hear audible differences at different sample rates.
And if you manipulate the audio material afterwards, you can also hear differences.

 

Last edited: Dec 3, 2019

Funny, but there is something to it when it comes to variable sample rate samplers [like Akai S-950, Mirage, E-MU II] which didn't resample audio to 44.1 or 48kHz while playing those samples back, all they used is a nice, gentle [or less gentle] LPF on the output. I used to sample everything at the highest sample rate into my S-950, well - audio with lots of highs, or audio that I played back transposed up to an octave or even two. Samplers with variable sample rate are simply great for transposition of audio. Nowadays I use TAL-Sampler in variable rate mode for that. Read about it in the manual for more info.

 

If you use SRC like R8Brain, iZotope, or SoX, you don't lose any quality. These are perfect. These days most of the software caught up with SRC quality, even Steinberg does decent SRC conversion these days. [It used to be terrible to the extreme in Wavelab and Cubase].

 

This plugin is useless in relation to anti-aliasing. Since aliasing, which is already included in the material, cannot be removed afterwards with a FIR filter. For this to work, the plugin would need to oversample before aliasing artifacts occur, not after.

 

No. 96kHz has enough audio bandwidth - the nyquist frequency is high enough so that the frequencies that reflect back [aliasing] into audible bandwidth are too low to be heard.

 

Last edited: Dec 3, 2019

Agree, but the question is can you hear the noise @ -100dBFS or greater?