16-bit... It really is enough!

[rkbates]

I totally agree, most of my music collection was ripped from CDs to hard drive. Played back on my current set-up, which is without a doubt the best and most revealing I've had to date, some albums sound fantastic and have scaled with the improvements in my playback equipment. Whereas some albums were so poorly recorded in the first place that they sound little better than they did on cassette played on the crummy Walkman and headphones I had as a teenager - nostalgia and rose tinted glasses notwithstanding.

One of the albums I bought on Bandcamp recently came with two copies of each track, one 16bit 44.1khz, one 24bit 88.2khz, I thought it would be interesting to see if I could hear the difference. To my surprise, I could, but if Iook at the file information it looks as though something else is going on, the 24bit versions of the tracks have a higher dynamic range and when played they are slightly quieter than the 16bit version:

View attachment 122979

Conspiracy theory alert - is the 16 bit file being deliberately degraded to make the 24 bit file more attractive and push consumers to pay the premium? Is that really a 6 dB dynamic range on some songs!

 

[Frank Dernie]

Reiss, J. D. 2016. A Meta-Analysis of High Resolution Audio Perceptual Evaluation. JAES 64(6): 364-379.


http://www.aes.org/e-lib/browse.cfm?elib=18296

Oh dear, thank goodness I haven't had extensive training

 

[Count Arthur]

Conspiracy theory alert - is the 16 bit file being deliberately degraded to make the 24 bit file more attractive and push consumers to pay the premium? Is that really a 6 dB dynamic range on some songs!

Regards the conspiracy, given that I bought the album as a download and had no idea that it contained two copies of each track beforehand, I doubt it.

I bought the album at Bandcamp: https://bandcamp.com/, I've written about this before, but once you buy an album or track from them, you can download it in a number of different formats, MP3, FLAC, WAV, etc.. Several albums I've bought have been higher rate than 16bit 44.1khz, but no mention of this was made on the site and I only discovered it once I'd downloaded the file, which was a nice bonus.

Unfortunately, I don't have any other tracks on hand in multiple bit rates to compare. Regardless, given the choice of a higher resolution version at no extra cost, I'll take it. The extra download time and storage space required really isn't an issue these days.

 

[Wes]

Maybe I'm wrong, but is your point that this study is solid evidence for the necessity of > 16 bit audio?

Reiss did the most careful study on this topic that I am aware of. It is a meta-analysis.

Conclusions indicate that while a difference does exist, it is fairly small.

As a matter of common sense, I preserve any HiRes material I have in its native (Hi Res) format. Storage is cheap.

I down convert it using XLD to Apple Lossless and play that.

There is no necessity for any audio.

 

[Sal1950]

As a matter of common sense, I preserve any HiRes material I have in its native (Hi Res) format. Storage is cheap.

I down convert it using XLD to Apple Lossless and play that.

You've lost me there, why would you archive the HiRez file but downsample it for listening playback, it doesn't wear out?

 

[Blumlein 88]

You've lost me there, why would you archive the HiRez file but downsample it for listening playback, it doesn't wear out?

I wondered the same thing. But if he listens over a phone or portable player the reduction in space may mean something or maybe his car audio doesn't do hirez. That sort of thing.

 

[Wes]

you archive the HiRez file in the (unlikely) event that some new innovative item of gear makes a difference audible, either to you or to an heir

 

[nimar]

It's been touched on but not explicitly stated, would the only reason to listen to a 24/48 or 24/96 track be that it was originally recorded at 24/96 and as a result the 16/44.1 would suffer from quantisation noise. Or is dithering good enough that this would make no discernible difference.

I switched Tidal to Hifi, instead of Master in Roon and there are many cases where both a 24/48+ (generally MQA) version exists and and a 16/44 version. But as many people record at variables of 48khz the 16/44 version would be inferior? You'd ideally want a 16/48 version.

ps. I've started down sampling to 44/48 for local hi-res files I own. Roon is unimpressed, green dot for downsampling.

 

[ebslo]

How do you figure? Lossy compression is better termed perceptual coding since uses a model of human perception to encode only that which can actually be perceived, discarding the rest. Some perceptual codecs are quite advanced indeed, taking advantage of psychoacoustic effects such as masking to avoid coding unnecessary information. Now if you have a recording sampled in 24 bits at 96 kHz, what is the most obvious way to reduce the amount of information without affecting perception? Why, remove everything above 20 kHz of course. Then remove anything below about -100 dB. These parts are uncontroversially inaudible, and besides, most playback systems can't reproduce them accurately, if at all, anyway.

Alternatively, suppose you want to do a lossy encode of a 24/96 file without discarding (all) the high-frequency content. How do you choose what to keep above 20 kHz when the perceptual model says none of it is audible? What do you throw out instead to make room for it?

For > 20kHz I agree the result would be the same, but for bit depth this is not true. Consider how people have been able to ABX 16 vs 24-bit. They find an almost silent part of the recording and crank the volume until they can hear the difference in noise floor. But a compression algorithm would throw out high-order bits that are all 0 for a block in a quiet section, instead of low-order bits, preserving a much lower apparent noise floor. This simple difference would likely be enough to make high-bitrate lossily-compressed content indistinguishable from the original 24-bit master, whereas the 16-bit version is distinguishable.

 

[fieldcar]

FYI. Someone asked about what dithering adds to the noise floor, it varies based on the resampler. Some really suck with tons of aliasing/harmonics, and many of the good ones sit around -140dB, and the best ones are about -160dB. Change the Test result to 1KHz tone to see the FFT.

https://src.infinitewave.ca/

 

[mansr]

For > 20kHz I agree the result would be the same, but for bit depth this is not true. Consider how people have been able to ABX 16 vs 24-bit. They find an almost silent part of the recording and crank the volume until they can hear the difference in noise floor. But a compression algorithm would throw out high-order bits that are all 0 for a block in a quiet section, instead of low-order bits, preserving a much lower apparent noise floor. This simple difference would likely be enough to make high-bitrate lossily-compressed content indistinguishable from the original 24-bit master, whereas the 16-bit version is distinguishable.

Any compression algorithm worthy of the name will do that, lossless ones too.

 

[ebslo]

Any compression algorithm worthy of the name will do that, lossless ones too.

Of course, except that lossless algorithms don't throw out any bits. But the algorithm can only preserve what was in it's input, so if the content was chopped from 24 to 16-bit before being compressed there is no way for the compression to preserve any of the 8 lowest order bits of the original, they're already gone.

Edit: I see what you're getting at, high order bits will be recovered perfectly so they're not really thrown out.

 

[mansr]

Of course, except that lossless algorithms don't throw out any bits.

The very definition of compression is to store information using fewer bits than the standard representation.

 

[ebslo]

The very definition of compression is to store information using fewer bits than the standard representation.

Ok, but I don't see what that has to do anything I was saying. We seem to be talking past each other...

 

[Frgirard]

Of course, except that lossless algorithms don't throw out any bits. But the algorithm can only preserve what was in it's input, so if the content was chopped from 24 to 16-bit before being compressed there is no way for the compression to preserve any of the 8 lowest order bits of the original, they're already gone.

Edit: I see what you're getting at, high order bits will be recovered perfectly so they're not really thrown out.

The lowest bit on cd 16bit are noise. The 24 bit adds 8 bit of noise. .
If you know a cd with an amplitude of 70 dB, i will be happy to know him.

50 dB it's unbearable in apartments, head phone, car, shuttle, crew dragon and ISS.
And no speakers can reproduce that

 

[Kegemusha]

A good article here.

 

[Wes]

not good - he should have put the phrase not a DBT up front instead of buried at the end to save people's time

 

[welsh]

Currently listening to the old 80s CD of Sting's Nothing Like the Sun. Fabulous recording and wonderful sounding CD. The CD has excellent dynamic range and even in the quietest parts, there's absolutely zero audible hiss.

The basic point is that 16-bits provides massive dynamic range. Unless you're hearing the quantization noise from your 16-bit digital source, there's zero reason to need higher bit depth. I guess it is possible to hear this quantization noise with headphones blasting and the right source material. With loudspeakers, I cannot see how you'd ever hear this quantization noise.

In addition, harmonic distortion can be almost completely eliminated with dithering, so really lower quantization noise level is really the only benefit of higher bit depth.

If you're not hearing quantization noise hiss, HIGHER BIT DEPTH DOES NOTHING!!!

I'll stick with my 16-bit/44.1KHz bliss.

There are three situations with a recording: 1) The format: Are ‘Hi-Res‘ formats better than the 16-bit standard that Philips et al originally identified? 2) The recording space, where the microphones were placed, how it was mixed, etc. 3) The actual performance of the musicians. I have six recordings of Schubert’s (very late) string quintet, and, musically, the best is a 1930s mono masterpiece by The Hollywood String Quartet (With added cello). No ‘audiophile’ would play this scratchy but brilliant interpretation.

 

[welsh]

And possible cartridge mis-tracking though not to the level of Tchaikovsky.

My father has some 1920s Gramophone magazines. Big topic - thorn or steel needle for Wagner? Yup, you guessed it - steel all the way!

 

[BrEpBrEpBrEpBrEp]

Dynamic range is absolutely critical. I actually require 32-bit audio for anything I listen to. Objectivists will sneer, but they just don't understand the magic of hearing someone halfway across the city breathe, the instant before a snare drum puts you in a permanent coma.