The Great 16 Bit vs 24 Bit Challenge

Waveforms which sound identical can look quite different on a waveform display. My hope was to make them look different enough so that people could not simply inspect the files to gain foreknowledge which inevitably creates bias. The flaw was that it was possible to attribute the copies to one of two parent files. This would have been avoided if I had had the foresight to have asked Steve to make multiple recordings at each bit depth. The difficulty with doing a test like this over the internet is that it requires the sharing of downloadable files. There is no way to stop individuals from looking for clues. The goal for this experiment was to have a blind test based on listening alone with no knowledge other than that provided by one's ears.

Joe Hardy

Waveforms which sound identical can look quite different on a waveform display.

Yes, interesting how that works. One place to look for differences on a Spectral Analysis (NOT involving high frequencies) is in the low-frequency Thunk (against a silent background) before each of the four sections, representing a Piston Press to set the registration for that section, and combining all the appropriate stop-knob noises in the Sample Set. Visibly, they all look different, file to file. Maybe if they sound a little different too, that could throw people off the scent, because it doesn't correlate with Anything... least of all Bit-Depth of the Sample.

The goal for this experiment was to have a blind test based on listening alone with no knowledge other than that provided by one's ears.

In this, I believe you have succeeded admirably. Thanks again for the challenge and the interesting discussions.

I enjoyed reading about digital to audio conversion on Wikipedia:
http://en.wikipedia.org/wiki/Digital-to ... _converter
It's clear there is a great deal going on here, "behind the scenes".

-- OrganoPleno

Audio expert Ethan Winer has created a very similar challenge that you can try. He has five downloadable files from 24 bits down to 9 bits.

Your challenge is to download the five Wave files linked below, and see if you can identify which file has what resolution. But you have to tell which is which by listening to them. No fair looking at the bits in an editor program! Each of the Wave files below are just over 2.2 MB in size.

Ethan's article is here:
http://www.ethanwiner.com/BitsTest.html

Joe Hardy

Revisitingthis thread because tnteresting feature on BBC Radio 4 earlier today about high resolution recordings. They featured a clip of Elgar played by the BBC Phil in 16 bit and 24 bit sound. They asked the recording engineer who recorded the clip, a recording producer and the presenter, using high quality listening equipment, to decide which was 16 bit and which was 24 bit.

All three were wrong.

http://www.bbc.co.uk/programmes/b03wq2sr

http://downloads.bbc.co.uk/podcasts/rad ... _yours.wav

if there is any difference it is only a very subtle difference when I play it on my Hauptwerk PC and I can't discount observer bias having heard the result on a standard radio before listening again on my professional setup.

I know there's been a lot of back and forth on this one and the subject is always of interest to me. I know the general consensus is there is no difference and not even the most trained ear can tell the difference between 16 and 24 bit, however I had somewhat of an oddity the other evening and am now again wondering if I'm hearing things or perhaps some of it has to do with the way the recording is done. I was doing some live recording with my handheld Zoom H1 recorder, it is capable of recording in both .wav and mp3 format in both 16 and 24 bit. I made the first recording in .wav format @ 16 bit, plugged in my headphones to listen to the playback and was quite disappointed with the quality, it had a considerable amount of hiss in the background, so I redid the recording in mp3 format @ 24 bit and it was remarkably much cleaner with better clarity and zero hiss. Was it the format that made the difference or the bit rate, or both?

Marc

I think it is important to realize that Hauptwerk is not playing a single recording, but (with large sample sets) easily a thousand recordings at a time. The noise of these recordings accumulates and is in some cases clearly audible as a hiss.

I think it is important to realize that Hauptwerk is not playing a single recording, but (with large sample sets) easily a thousand recordings at a time. The noise of these recordings accumulates and is in some cases clearly audible as a hiss.

Quite true, but it also true of wind noise which is invariably much greater than the -96 dB quantization noise of 16 bit audio. This wind noise is inherent in all pipe organ samples (even after judicious digital noise reduction) and it completely masks the quantization noise.

Joe Hardy

Hi Joe, very good point. But I think that wind noise is limited to certain frequencies, while quantization noise accumulates at all frequencies.

Hi Joe, very good point. But I think that wind noise is limited to certain frequencies, while quantization noise accumulates at all frequencies.

Thank you, josq. Wind noise and quantization noise are both broad band similar to white noise. The single biggest headache for sample set producers is wind noise for exactly the reason you cited. The original pipe organ may have been reasonably quiet and the blower noise does not change when more pipes are added. In sampled pipes, blower noise is unavoidably recorded along with each individual pipe. I personally have recorded and processed many thousands of pipes over the last twenty years and have never found any where the noise level was even -60 dB (about the same as 11 bits). A pipe in a loud rank will mask the noise while it is sustaining, but the noise becomes obvious as the reverb tail decays. As you pointed out, when lots off virtual pipes are layered, the noise is additive.

Joe Hardy

Thank you for the explanation. I was convinced of the effect because I diagnosed it in the Krzeszow set with my previous setup. Today I did the test again -different room, different speakers - and I was not able to hear significant differences.

So I'm not convinced anymore that quantization noise at 16 bit quality is a significant effect in Hauptwerk. I think you provided good arguments.

I have to correct my previous post. I actually COULD reproduce the effect in the Krzeszow set, when playing "full organ". During the day, I could barely notice it because there is quite a bit of background (traffic) noise in my appartment. But in the quiet evening hours, clearly a hiss is present in the release tails when played in 16 bit quality, which is absent at 20 bit. Once you notice it, it is on the brink of being disturbing.

If interested, I can send you the recordings.

First of all - many thanks to Joe Hardy for contributing such an interesting issue (once more ) - and the trouble he has gone through to prove it.

I´m not capable of contributing to this thread from my own knowledge - whether concerning the theoretical or listening issues.
Besides - listening to these clips, it just crossed my mind that when the recordings would have been made with a baroque organ and therefore would probably include more bright stops, would that have made a difference? Well, at least I think it would have helped me...

Anyway, my question really is the following.
I have great appreciation for the VPO sample sets made by Juri Zurek. He clearly advises not to use 16 bit but 20 bit: http://www.sonusparadisi.cz/en/blog/do- ... in-16-bit/.
He has made quite an effort to illustrate his point.
Again, I can´t judge this myself by listening, and besides I´m an absolute 24 bits freak, but it strikes me that no one has referred to this statement yet of someone who has earned a solid reputation in the field of VPO sample sets.

Juri himself has not reacted to this thread up till now, but as I regard him as an authority in this area, I hope he does not mind that I put his (public) opinion as an issue in this conversation.

Ernst

I too have read and followed Jiri's directions / findings to load his sets in at least 20 bit (and have done so) and would be interested to hear more detail about his reasoning for this. As I concluded here earlier, in my own recording experience the optimal bit rate to load a set in perhaps has something to do with the bit rate used to record the set in the first place? i.e., set recorded in 24 bit, load it in 16 bit and you will hear a difference, set recorded in 16 bit and loaded in 24 bit and you won't hear a difference? This is all just a guess on my part for now and I still may be just imagining that I'm hearing things.

Marc

Hi Marc,

Almost all recent organs (Hauptwerk vs. 3 & 4) are offered in 24bit/48kHz but have been recorded in 96 kHz.
So really the issue - and Juri´s opinion - has to do with the question: when a VPO is offered in 24bit/48kHz , can you hear a difference when it is loaded in 16 bits/48kHz instead.

Ernst

and Juri´s opinion - has to do with the question: when a VPO is offered in 24bit/48kHz , can you hear a difference when it is loaded in 16 bits/48kHz instead.

Ernst

Hi Ernst,

At least according to what I've read on Jiri's site, he says there is a difference. Without visiting his site directly, I can't remember his exact words but do recall him writing about a certain set of his and going to the extent of saying that by loading it in 16 bit rendered the set unplayable in his opinion. Now I can't say that my ears are that sensitive and have never tested to verify, but who knows? What's unplayable to one might not be to another?

Marc