Μαθημα ηχοληψίας #1

[gerasimos]

1) Εχω κάρτα Edirol UA25 24bit/96k' date=' εχει νόημα που εχω το nuendo να γραφει το project στα 32bit? Ποια ρυθμιση να προτιμήσω?? [/quote']

 

1) Το 32 bit του nuendo αναφέρεται στην εσωτερική ανάλυση του προγράμματος. Οπότε για να αποφύγεις ένα dithering παραπάνω θα σου πρότεινα 24 bit.

 

Ναι και όχι. Τα 32 bit είναι όντως η εσωτερική ανάλυση του audio engine του Nuendo' date=' δηλαδή το βάθος bit με το οποίο επεξεργάζεται το κάθε sample ενός track. Δεν μπορείς όμως να πεις ότι "γλυτώνει ένα dithering", για τον εξής λόγο: όταν ένα plug-in εφαρμοστεί σε μία ακολουθία από 24-bit samples, τότε προκύπτουν αριθμοί μεγαλύτερης έκτασης των 24 bits (ιδίως σε συστήματα κινητής υποδιαστολής όπως το Nuendo), και αυτό που συμβαίνει είναι αλλεπάληλα ditherings προκειμένου να κβαντιστούν τα samples και πάλι στα 24.

 

Από την άλλη, πρέπει να ξέρεις αν τα 24 και τα 32 που αναφέρεις είναι σταθερής (fixed) ή κινητής (floating) υποδιαστολής. Για παράδειγμα, τα 32-bit floating point έχουν ουσιαστικά 24-bit πληροφορία (τα άλλα 8-bit είναι ο εκθέτης), και ακουστικά σχεδόν δεν υπάρχει διαφορά στα δύο format (32-bit floating με 24-bit fixed).

 

Παραθέτω ένα αρκετά ενδιαφέρον κείμενο του Paul Frindle (επικεφαλής της ομάδας ανάπτυξης των Sony Oxford plug-ins):

It is important to realise that errors creep in and are added at every point of processing - addition - multiplication whatever - you don't get something for nothing in digital any more than analogue - its just that digits being numerical can be made to any accuracy - analogue unfortunately has constraints imposed by the physical world. Anyway every resulting signal after an operation of any kind (apart from one that changes nothing) Must be regarded as an entirely new signal with the accuracy of the function that got used. The operations have no knowledge of what preceded them! So for instance if you change the level of a 24bit signal (i.e. multiply it) by anything other than a whole number of power of 2 levels i.e. 6dB steps) you will end up with a signal that requires more precision than 24bits to represent it - regardless of it having been a perfectly dithered 24bit signal in the first place. So the accuracy of the internal math determines how much degradation there is likely to be. So for a 48bit accurate multiplier (like in PT) the result of changing the level of a 24bit signal is in fact a new 48bit number (cos of the multiplier precision). The trick is to keep that 48bit data path intact until the final output and then turn it into the required 24bit output by re-dithering it. In this case the error contribution from the mixer will be at -288dB and we dont care about it because as Nika says there will be enough noise to dither the harmonic quantisation errors out anyway and we can't hear at 288dBs down.

Thats fine until we go in and out of a 24bit port like the plug-in interfaces. These force the data to be represented at 24bits, so the maximum accuracy we can get is a new dithered 24bit signal at every interface (or harmonic noise at -140dB or so if it isn't dithered). However if dither is properly used this still doesn't cause too much hassle because the (uncorellated) noise content adds at only 3dB for every doubling of the times the data is truncated to 24bits and added together. So in practice getting noise worse than -120dB or so is actually quite difficult and everything should still be good enough for rock and roll - hell of a sight better than the 70dB or so dB we got from analogue tape anyway.

But there are other more practical issues that affect the platform that give people gip. One important issue is that the lack of headroom at the effect interfaces causes any increase in level to clip the signal (if you record flat out because you believe in resolution) and you end up turning down gain controls all the time to prevent overs. The result of this is often that something gets to be too quiet to gell in the mix and you have only 6dB of gain on the fader to get it back again - very tiresome.

Another horribly important issue is that the meters and the waveform displays tell you only sample value - not signal level! This is because the samples have not yet been re-constructed and decoded (its Pulse CODE modulation - right). Not only does this lead everyone to argue forever about 'resolution' cos thats what you apparently see on the screen (yawn), but worse still it does not actually tell you the resulting signal level after decoding. So you can have signals that dont turn on the over lights that will STILL actually clip at the output DAC!

Ok now onto the floating business. Floating data consists of a mantissa (the meaty bit) and the exponent (the scaling bit). So for instance a 24bit mantissa signal with an 8 bit exponent can represent a 24bit signal accuracy at vast ranges of values. In other words the data with basic accuracy can be maintained at 24bits (SNR of -144dB) whatever the actual signal value level happens to be within reason. So a signal at -100dBs can still be delivered at 24bit precision to the next function. This sounds like a great idea, but the only rub is that the base precision depends on the mantissa and not the exponent, so this example has a need to re-dither (or already have noise) at 24bits for each transfer - so its really no better than a 24bit fixed system in absolute terms and you've used up 32 data bits to get it. Using the 32 data signals as fixed point instead would have been much better.

BUT in the specific case where signals get larger and smaller indiscriminatly it actually fares much better because you don't need to lose math precision (SNR) to accomodate internal overload margins. You can for instance boost your EQ and the signal will not degrade by being too big, you can reduce the signal and it will not degrade by falling into excess truncation 'noise' due to math scaling errors or the original signal's dither falling off the bottom etc.

So although floating point is actually a theoretical disdvantage in a well designed system, for applications without any headroom margin with operating levels that coincide with absolute max modulation, with meters that don't read actual level and users that are obsessed with recording flat out (because they misunderstand 'resolution'), the floating format actually provides a real (if accidental) benefit under these awfully disadvantageous conditions. And some people may find that the practical business of mixing is actually inexplicably easier on some existing floating mixer applications, especially since they may in fact be using far greater floating precision than my example.

On the other hand, if you find yourself in front of one of the affected systems, you can do yourself a big favour by recording at lower levels (i.e. -12 to -20dB - if your plug-ins are of high enough quality), take no notice of the quantisation of the wave display (please!) and aim for a final mix modulation level of around -1 to -2dBs below what the meter says is flat out, even if that does mean you need to insert some gain before the main fader somehow.

[tugito]

H αλήθεια ειναι οτι ακομα δεν εχω καταλάβει καλά αν πρέπει να εχω το nuendo σε 32 η 24 bit. Αυτο που εχω καταλαβει ειναι οτι ειναι καλύτερα σε 24 λόγω processing και πραγματικών διαφορών μη αντιληπτών.

 

Οι κιθαρες στο ''Ελευθερα παιδια'' μου φανηκαν

πολυ καλες, (στα κουτακια του pc που τα λενε ηχεια)

 

Thanks! Ειναι ηχογραφημένες με το jhonson (ειναι μια καλή παλιά συσκευη για όσους την ξέρουν)

 

Παντως κατέληξα σε ήχο πολυ καλό για κιθάρες τελικά. θα ανεβάσω συντομα το κομματι να μου πείτε γνώμη. Εχει ντραμς ezdrummer το καθενα στην εξοδο του και δουλεμένο με κομπρεσορ και eq καθώς και λίγο ελάχιστο reverb στο ταμπουρο και στο hihat.

Επισης εχει μπασο απο broomstick bass το fender jazz fingered περασμενο απο ampeg ενισχυτη (vst) και φυσικά κομπρεσορ.

Τελος εχει κιθάρες που για να καταλήξω ακουγα διαρκως σε εναλλαγή το white people for peace - AGAINST ME! που ειναι ο ήχος που θα λάτρευα να έχω και κατεληξα στο ότι εχουν τις κιθαρες παναρισμένες με κάποιο τρόπο πίσω από το κεφάλι. Εβαλα λοιπόν το plugin πανόραμα και το πέτυχα. Επίσης εβαλα μια κιθαρα αριστερα να παίζει πιο μπασατα και μια δεξια με πιο μεσεα και ψηλα και πιο πανκ, οπως δηλαδη τα ακουγα στους agianst me. Να ναι καλα το amplitube 2 και το gtr2 και τελος τα εστελνα και προς ενα bus με το panorama και ειμαστε κομπλε.

Θα το ανεβάσω soon...