Rendering 16 or 24 bit?

[pedrocaetanos]

(...)96dB alone is probably more than some of the analog components behind the D/A playing your music can do and given that most music doesn't have more than 24dB of effective dynamic range its well enough for a final output.

I have to digress on this. Maybe true for most music done nowadays in the Pop genre, but if you hear styles like jazz and its derivatives, New age, and a bunch of world music styles (for example Portuguese fado, or for example norwegian folk inspired jazz - Jan Garbarek anyone?), you will have very dynamic stuff from 50 db up! Not to mention that many classical music concerts can have a wide dynamic range of 60 to 80 db's.

Most good monitors, and standard gear, made up to the 2000's (not hedious mp3 players and yucky soundbars) and still today's good hi-fi stuff will have sensitivities of -80 to -90 dbv from unity.
Of course for each component you put on the chain you will take out some sensitivity, but at max, most complete systems have a -80 db sensitivity, and that's A LOT!

I do have some Jan Garbarek CDs
Don't remember if the DR of those is high or not. But I doubt fado has a DR of 60-80 db's.
Classical music, that's another story. But even so, probably most performances, if well recorded and well mastered would fit in the standard 96dB of DR.
And there will always be exceptions that will benefit with the extra DR. In several music genres.

[selig]

My simple rule - if exporting for further processing, bounce to 24 bit. If exporting as a final master, bounce 16 bit and add dither.

[mcatalao]

I do have some Jan Garbarek CDs
Don't remember if the DR of those is high or not. But I doubt fado has a DR of 60-80 db's.
Classical music, that's another story. But even so, probably most performances, if well recorded and well mastered would fit in the standard 96dB of DR.
And there will always be exceptions that will benefit with the extra DR. In several music genres.

Jan Garbarek's cd's are quite dinamic. Fado is also very dynamic, now that you got me thinking about it, there are quite soft passages, but maybe 50 db was exagerated?

But 20 db seems to be too low nonetheless. The thing is you have to see this in what it really affects. I am no DSP mogul, but in this ranges if you go from 16 bit to 8, you will listen the quality loss (going from 96 to 48 db on a fado performance, or in those Garbareck CD's is noticeable!).
My point is that the DR in the digital domain, affects quality at the sample level. Or else why use 16 bit files when we could compress the hell out of music, and use only the top 4 bit of our songs?

I do not want to start any bit and byte war, and i'll be happy to rest the case with the same opinion as Seligs, wich is how i've been working till now.

[mcatalao]

My simple rule - if exporting for further processing, bounce to 24 bit. If exporting as a final master, bounce 16 bit and add dither.

Hi Selig,

Have you had any "Master for itunes" loss-less request, and if so, are you starting the recording at more than 44.1 kHz@16 bit?

I had a client asking for mp3, Itunes Lossless, and CD, and only asked me the losseless format AFTER recording the vocals, which we had done so in 44.1 kHz... I guess it's not the same, but i end up mixing everything at 88.2 kHz@24 bit then mixed down to the different formats... :/ He didn't complain, so i don't believe he noticed!

[selig]

My simple rule - if exporting for further processing, bounce to 24 bit. If exporting as a final master, bounce 16 bit and add dither.

Hi Selig,

Have you had any "Master for itunes" loss-less request, and if so, are you starting the recording at more than 44.1 kHz@16 bit?

I had a client asking for mp3, Itunes Lossless, and CD, and only asked me the losseless format AFTER recording the vocals, which we had done so in 44.1 kHz... I guess it's not the same, but i end up mixing everything at 88.2 kHz@24 bit then mixed down to the different formats... :/ He didn't complain, so i don't believe he noticed!

As I said, 24 bit for EVERYTHING except the final master, ESPECIALLY when tracking audio.

As for sample rates, I've been recording at 44.1 kHz in DAWs since the beginning, with some work at 48 kHz for video destinations. But for soft synths and IDB mixing, there are some who find higher sample rates to achieve certain things for them, none of them being "extended frequency response" of the final master, and most of whom deliver at lower sample rates in the end anyway. The higher rates are mostly chosen to avoid aliasing by moving nyquist (and therefore artifacts from aliasing) up beyond the audible range, rather than to achieve some sort of higher bandwidth or affect the mysterious "resolution" parameter…(whatever THAT is).

I'm of the school of thought that minor issues are unconsciously adapted/compensated for by most engineers - if you give them dull monitors or have them record to tape that changes the sound on playback, they will quickly adapt and adjust their workflow to compensate! So in the end, minor differences between sample rates and bit depths are often worked around with EQ/filtering etc. where possible and in the end the results are often indistinguishable from each other. In other words, give a Samurai a kitchen knife or a Samurai sword, and either way they will manage to win the fight!

[pedrocaetanos]

Fado is also very dynamic, now that you got me thinking about it, there are quite soft passages, but maybe 50 db was exagerated?

"maybe"...

My point is that the DR in the digital domain, affects quality at the sample level. Or else why use 16 bit files when we could compress the hell out of music, and use only the top 4 bit of our songs?

In basic aspects, you are obviouslly right. But after a certain number of bits do humans really hear the difference? The discussion always revolve around that. Often there is much disagreement between the points of view of of the two extremes: enginers, and audiophiles solely based on subjective audiotioning (usually there's also esoteric hi-fi G.A.S. involved...)

I do not want to start any bit and byte war, and i'll be happy to rest the case with the same opinion as Seligs, wich is how i've been working till now.

Neither do I, and I also think Selig summed it up very pragmatically

[normen]

I do not want to start any bit and byte war, and i'll be happy to rest the case with the same opinion as Seligs, wich is how i've been working till now.

There is no case. I said the same thing about exporting 24bit for further processing and 16bit being enough for a final export. You just dwell on the fact that I said that most music has an effective DR of not more than 24dB - which is true if you don't count the 0.5 seconds of reverb dying at the end of the song, which isn't really an important part of the music, is it

[dvdrtldg]

I make radio documentaries and record at 24/48, output at 16/48.

48 simply because that's what we use at work, i cant hear any difference compared to 44.1. But 24 for recording makes sense if you're using the good gear

[Benedict]

Slightly OT but related closely enough.

For the last year of so my albums have been composed and rendered at 16/44. They always sound the way I expect. If there is a nyquist event I hear and make decisions at the start.

I tried this the other day:

1) make a Thor patch with Osc 2 Mod Osc 1 (Analog Sine, not FM Mod) till I had lots of nyquistian activity. I rendered at 16/44 and played in Win Media. Sounds exactly as I expect.
2) switched Reason to 96 and the sound smooths out, dramatically less nyquistionation.
3) exported from Reason at 16/44 and listened in Win Media. It sounds just like the first render with all the nyquisteny back
4) exported at 24/96 and listened in WinMedia (which now plays that file size) and it sounded like it did at #2, nice and smooth
5) I used Voxengo r8brain to convert the 24/96 to 16/44 and the new file sounds almost exactly like #2 & #4

This tells me that there is no right way to work. If it sounds good it is. If I want the extra smoothness (reduced nyquistivity) in my sounds then I take the hit and can't render from Reason at 16/44 as that is the rate it renders at and is not really rendering at 24/96 and downsampling. r8brain is a free and pretty easy solution, esp now that WinMedia plays 24/96 files without complaint so I can listen to album tracks without lots of dummy renders.



yes I know Aliasing is the term but that is far less fun that making up fancy terms. besides if you know I am wrong then there is no need for me to be right as you done gotted me first right

[ScuzzyEye]

Here's something that might twist a few minds.

If you render audio at 96 kHz, and then brickwall filter it so there is no spectral energy at or above 24 kHz, you can then simply drop every-other sample to make a 48 kHz file that will sound the same as the 96 kHz. In fact if you try to get fancy and average the two samples together into one, you'll likely introduce inharmonic artifacts.

[Bonkhead]

44.1 at 24 bit

[satyr32]

Blind Test regarding mp3 & wav. Even there the differences are hard for me to spot.
http://www.npr.org/sections/therecord/2 ... io-quality

[pjeudy]

The site with the graphs was probably this one: http://src.infinitewave.ca/

It's always interesting to learn the technical aspect of sounds, but I will admit that I don't understand it all...can you shed some light as to what's good/bad regarding this comparison Ableton and R7

[selig]

The site with the graphs was probably this one: http://src.infinitewave.ca/

It's always interesting to learn the technical aspect of sounds, but I will admit that I don't understand it all...can you shed some light as to what's good/bad regarding this comparison Ableton and R7

Most importantly as it pertains to these specific graphics, you would ONLY hear this (if at all) after performing a sample rate conversion, since that is all the graph is showing.

[normen]

It's always interesting to learn the technical aspect of sounds, but I will admit that I don't understand it all...can you shed some light as to what's good/bad regarding this comparison Ableton and R7

The line going up is the actual frequencies that are in the audio, the colorful line going down is frequencies above nyquist (1/2 sampling rate) that get mirrored down into the audible range when the audio is being downsampled (at a higher sampling rate that would be the frequencies going further up). This is the effect called "aliasing". Ideally it shouldn't be there. That said, as you see by the color all of it is below -60dB so it shouldn't be too troublesome but as I said Reasons downsampling process isn't exactly one of the best out there so best work with a consistent sample rate in Reason.

As to how aliasing happens, if you consider this image, the red sinus wave would be a frequency that is too high to be captured by a certain sampling rate, the long wave is what is recognized instead (and what makes the colorful line). To counter this a steep filter at nyquist is applied to get rid of these higher frequencies so that only frequencies that are below nyquist are sampled. Very steep filters are computing intense so Reason sticks with a less steep filter than others, resulting in this low-volume aliasing visible in the plot.

[pjeudy]

Most importantly as it pertains to these specific graphics, you would ONLY hear this (if at all) after performing a sample rate conversion, since that is all the graph is showing.

The line going up is the actual frequencies that are in the audio, the colorful line going down is frequencies above nyquist (1/2 sampling rate) that get mirrored down into the audible range when the audio is being downsampled (at a higher sampling rate that would be the frequencies going further up). This is the effect called "aliasing". Ideally it shouldn't be there. That said, as you see by the color all of it is below -60dB so it shouldn't be too troublesome but as I said Reasons downsampling process isn't exactly one of the best out there so best work with a consistent sample rate in Reason.

Thank you both !
Ah...Thanks Norm that chart you added helped a little more!

[ScuzzyEye]

Thank you both !
Ah...Thanks Norm that chart you added helped a little more!

Everyone said everything I would have said.

Did you by any chance compare Reason 6.x to 7? Reason 7's results are entirely reasonable, and will likely only be audible in test signals. But Reason 6 had...issues.

EDIT: And if you're looking through the results, and see a program called Brick, and are wondering what it is (because it is the best in the list), it's here: http://camil.music.illinois.edu/software/brick/ It has a simple GUI in the Mac, and needs to be compiled for command-line use anywhere else. So not the most friendly program, and I've only used it as an experiment. I'm entirely fine trading a bit of ease of use for the results that Audacity delivers. But if you want that ten-tenths, Brick is your tool.