Agreed, Im calling bullshit on telling the difference in a blind test

I suggest everyone watch this video (relevant part starts at 41:50)

 

44.1/48 kHz
24 bits (for dynamic)

It is like in photography 14-16 bits RAW format : no sensor can fill it for now

 

Thanks, this is a fantastic share to pass around

 

most people get get confused by digital conversion because it is somewhat abstract.

Sample rate is about frequency range of the recording. 44.1k is fine but nowadays irrelevant unless you are going to burn straight to cd. This is because when the CD was invented the only storage big enough for professional transfer were digital tape storage and the maximum sample rate they could fit which would give a 20-20k range was 44.1k (the tapes were similar to vis which striped diagonally across the tape.)
44.1k is an awkward number in pure digital storage as it doesn't divide easily which is why 48k became the standard across digital video (the next development) and which is why most audio is a variant of this as film/video is big business. so 48x2 =96, 96x 2= 196 etc

the nyquist filter was designed to stop harmonics (which can be vastly louder that the original single) driving the digital signal into distortion. A good analogue system could cope with over-modding ad unlike digital, analogue distortion is relative to the overpeak, i.e. a peak would distort in the same shape and time (often used to as compression), unlike digital which simply distorts across all frequencies which souls awful.

Sample rate is a bit like colour in digital video, a few 1000 more shades is not really noticeable unless you have a very, very good sound system.

Bit rate on the other hand is the resolution, the pixels if you like and is crucial, particularly in live recording as the more you have the more you can leave a safe headroom and still have a detailed recording.

the key to a good system is the accuracy of the converters from analogue ( such as a mic) to digital (your DAW)

analogue means literally analogous to, ie an exact copy. so an analogue sine wave is a true curvy sound wave. A digital conversion breaks that down into steps and however small they are, each step is not a curve so it produces a small error which we perceive as noise. This noise is not relative to the signal and so can be a problem. There are different ways to disguise this noise (dither, noise relative to signal is added)
but essentially the quality of this conversion matters. This is why mastering houses may spend $4K + on a two channel a-to-d converter.

The key to keeping quality in the digital world is make sure all your sample rates match, never use MP3 unless its artistically important and try to record at the highest bit rate you can

 

Create a 1kHz sine at 44100 and 192000. Then see waveforms are different, zoom in. 44100 will show low precision, steps... 192000 will show higher precision, more smoother.... BUT now hear them, they are both clean sines! No difference, you will not hear those steps.
And digital signal is reconstructed via DAC into sound you hear.

Digital is not synonymous for discrete.
Analog is not synonymous for continuous.

FabienTDR's words. Analog can be discrete too, digital can be continuous too. Read recent thread about hires audio on gearslutz mastering thread, he gives some detailed explanation about that.
I hope you know who is Fabien? =)

....
For me, higher bit depth - protection from quantization noise, more details for quieter parts of audio. Floating point protects from clipping above 0dBFS.
Higher sample rate - less aliasing, less cramping at 20000Hz for digital eqs, too subtle compressor's timing precision (attacks, lookahead), better re-pitching quality. Usually, 96k is not enough sometimes, so additional oversampling 2x or 4x may solve some problems. It is better to 2x oversample at 96k, than 4x oversample at 48k....less cramping for frequency/phase (if oversampling algos are bad). Or use 192k, but not all plugins support 192k...and not all interfaces support 192k...

SRC is your go-to for resampling. Src.infinitewave.ca/

SoX, izotope rx, dbpoweramp, r8brain, SSRC, ez cd latest, final cd sharpmode, the brick (mac), weiss saracon, several others.

 

How do they do the blind test? in order to properly do this wouldn't someone have to start a mix with either 96k or 48k files and mix all the way through to the same respective renders? Does dithering from 96k to 48k count?

 

This is somewhat of a myth or a misconception

4:17:

 

Engineers think they are so special, with their suspenders and fancy hat.

 

TL;DR But I think this should be the best answer:

All the other stuff may be interesting: professional stuff over 1K bucks, and so on. But's off-topic taking the wild guess the OP doesn't wanna spend that shitload of money on an audio interface

 

One of the best explanations around still, everybody should watch this.

 

Hell yeah!

 

Because they technically were never there. There's nothing "between the dots" in a digital signal, so it can't have steps. It's literally the same, provided we bandlimit it to half the sample rate.



The band limit is important - if we can be sure there is nothing above a frequency (sample rate / 2), there is only one way to connect the dots - and it will result in *exactly* the same waveform as the analog one coming in. It will not be in steps, it will result in the exact same smooth waveform, all the way up to Nyquist frequency.

Here's 8-bit audio, sine wave at 10kHz. Do you see any steps?



The idea of staircase steps is plain wrong, every commercial ADC/DAC these days is a delta-sigma one, it's actually 1 bit and very high oversampling:

First of all - a reality check. Here's a 1kHz sine wave at -5dB, 16/44.1k:


The noise is at -120dB, ie barely audible. Once you dither it you get this:



Uniform noise at -135dB. Your microphone/preamp likely adds a lot more noise than this, and so do your speakers.

Remember 16-bit is 2000 technology, we're all working at 24 bit now. Let's see again before dithering:



and after:



Note even before dithering the noise peaks at -150dB, which is already lower than -135dB we had with dithered 16-bit. This is 15dB less noise than audio CD. Did you hear constant noise on your audio CDs that you're concerned about?

Keep in mind internal processing (= anything before you export your project) is 32-bit float, with a dynamic range of over 1500dB. Also see this:



The 2 bit recording is not worse because it didn't capture some part of the signal - the signal is the same, but there's more noise to it. The signal is the same quality in all bit depths here. The only difference is in the level of noise. The only thing you gain by recording at a higher bitrate is less noise. Do you hear noise in your system that makes you believe you need more bits? With 16-bit you get 96dB of dynamic range, and it increases to 144dB for 24-bit. Human ear can be assumed to hear from 0dB (silence) to 120dB (pain). With 24-bit audio you can play your audio loud enough to cause physical pain, and you won't be able to hear the quietest 24dB. The actual range you can hear is 120dB, or 20-bit. But you can't even hear that, a top-tier recording studio has background noise at 20dB, your location will likely have more, so the actual range you'll hear is even less.

With 16-bit audio, once you dither you get 120dB of effective dynamic range. It's already enough, 24-bit is overkill, anything above that is audiophoolery.

Sure, but oversampling is used in pretty much all decent plugins these days and running your whole session at 192k is a total waste. See FabFilter video on oversampling. You can work at 44.1k and have better results with less aliasing/cramping by using plugins that oversample properly.

 

maybe for teenagers. lol not for 64 year olds like me.