Head-Fi.org › Forums › Equipment Forums › Sound Science › Double blind test 128Kbps vs lossless? I'll be amazed if you can tell much difference
New Posts  All Forums:Forum Nav:

Double blind test 128Kbps vs lossless? I'll be amazed if you can tell much difference - Page 11

post #151 of 257

Unfortunately, the links from the original post no longer seem to lead to the files.  Does anyone have them?

 

I did have another listen to one of my own tracks with which I could once successfully distinguish 16 and 24 bit versions.  I had just finished working on it when I did the original blind test.  Now, a few years removed, I can’t confidently distinguish the two.  So I didn’t even bother to repeat the blind test.

 

The bottom line is this: even if I can’t easily tell the difference now, I know that with thorough attention, it is possible.  And how do I know that the day won’t come when there’ll be headphones or speakers that will make the differences more apparent?  So I’d hate to lose the 24 bit versions. 

 

With hardrive space growing ever larger and inexpensive, why not have the best quality files possible?  I’m pretty tired of getting really into an MP3 release, listening to it a lot, and then hitting that sonic wall that the grain and high frequency roll off of the compression entail.  And the more detailed my monitoring gets, the sooner I hit that point.

 

I do want to say one more thing about 16 and 24 bit files, since I’ve had a lot of past discussions on the topic.  The 96 db dynamic range of 16 bit files is sufficient for almost any recording.  But as Dan Lavry has said, converters don’t end up quite giving us all 16 bits.  Dropping to 15 or 14 useful bits could actually be pretty audible.  At the recording end of the equation, the extra dynamic room from 24 bits really adds flexibility to manipulate tracks.  So if the source file is 24 bits, getting it to 16 involves dithering.  Unfortunately, the noise shaping that goes along with that does alter the signal.  I know that it’s measurable.  And I know that human ears can pick it out too.

post #152 of 257
Quote:
Originally Posted by EllisDee View Post

Highly detailed music (Classical, Shpongle etc) benefit massivley from lossless in my opinion.

 

I could never listen to music I love in mp3 format unless I was desperate and had rubbish headphones and no amp.

Its because of the iPod generation that Hi-Fi's in stores are rubbish and modern music is so poorly produced. Becuase they can get away with it an no one can tell on there iPod headphones/MP3s or low quality modern HiFi MIDI sytems. LOL again in my humble opinion.


Hey hey, guess what the test file is? Classical!

 

So what were your ABX results?
 

Quote:
Originally Posted by nnotis View Post

Unfortunately, the links from the original post no longer seem to lead to the files.  Does anyone have them?

 

I did have another listen to one of my own tracks with which I could once successfully distinguish 16 and 24 bit versions.  I had just finished working on it when I did the original blind test.  Now, a few years removed, I can’t confidently distinguish the two.  So I didn’t even bother to repeat the blind test.

 

The bottom line is this: even if I can’t easily tell the difference now, I know that with thorough attention, it is possible.  And how do I know that the day won’t come when there’ll be headphones or speakers that will make the differences more apparent?  So I’d hate to lose the 24 bit versions. 

 

With hardrive space growing ever larger and inexpensive, why not have the best quality files possible?  I’m pretty tired of getting really into an MP3 release, listening to it a lot, and then hitting that sonic wall that the grain and high frequency roll off of the compression entail.  And the more detailed my monitoring gets, the sooner I hit that point.

 

I do want to say one more thing about 16 and 24 bit files, since I’ve had a lot of past discussions on the topic.  The 96 db dynamic range of 16 bit files is sufficient for almost any recording.  But as Dan Lavry has said, converters don’t end up quite giving us all 16 bits.  Dropping to 15 or 14 useful bits could actually be pretty audible.  At the recording end of the equation, the extra dynamic room from 24 bits really adds flexibility to manipulate tracks.  So if the source file is 24 bits, getting it to 16 involves dithering.  Unfortunately, the noise shaping that goes along with that does alter the signal.  I know that it’s measurable.  And I know that human ears can pick it out too.


I uploaded it to Mediafire some time ago.

 

Also, I have a hard time believing you can distinguish 24 and 16 bit files of the same master. Did you downsample the 24 bit file to 16 bit (with dither), or did you use a different 16 bit source? There was another discussion thread where it was concluded most (if not all) of the audible difference is because of the mastering, not the bit depth.

post #153 of 257
Quote:
Originally Posted by Head Injury View Post

Also, I have a hard time believing you can distinguish 24 and 16 bit files of the same master. Did you downsample the 24 bit file to 16 bit (with dither), or did you use a different 16 bit source? There was another discussion thread where it was concluded most (if not all) of the audible difference is because of the mastering, not the bit depth.


Are you sure everyone that heard 24 bit actually can hear 24 bit?  As in they have supporting hardware?  Maybe I'm wrong but my DVD Audio on my XFi card is quite impressive over the same 16 bit master.  There are also issues from what people (industry folk) have called 20 bit and 24 bit remasters not being true 24 bit.

post #154 of 257
Quote:
Originally Posted by Anaxilus View Post

Are you sure everyone that heard 24 bit actually can hear 24 bit?  As in they have supporting hardware?  Maybe I'm wrong but my DVD Audio on my XFi card is quite impressive over the same 16 bit master.  There are also issues from what people (industry folk) have called 20 bit and 24 bit remasters not being true 24 bit.


Maybe. I'm mostly thinking of SACD vs. CD though.

 

And I've no proof that he can't hear it. But I doubt he can until he shows proof of his own.

post #155 of 257
Quote:
Originally Posted by nnotis View Post

Unfortunately, the links from the original post no longer seem to lead to the files.  Does anyone have them?

 


The original test files are still there.  I just downloaded them again to be sure.

 

You need javascript enabled.  Enter the captcha code.  Wait about 45 seconds for the countdown timer to get to zero.  Click the orange "Regular Download" button.  Their business model is to try to get people to pay for more convenient downloading without the restrictions.  So the free downloading gets encumbered to make it less convenient.

post #156 of 257

24 bit only, mathematically, has a lower noise floor than 16 bit.  You'd have to listen at 120db or louder to hear a difference, which I'm sure nobody wants to damage their hearing permanently to do.  Because of the resolution of the gear I'm using, I'll take 24/88.2 or 24/96 files when they are available, but last night after an upgrade in my system I was still finding new things I hadn't heard before in 16/44.1 (CD quality) files, so I'm not going to pretend that I'm really gaining that much from the higher bit-rate files.

post #157 of 257

I don't think thats right to say its just noise floor and you have to listen at 120db.  24bit is a lot more resolution meaning there is way more data in that file compared to a 16 bit, hence using DVD's over CD's.  That has nothing to do w/ noise.  Its not like MP3 encoding either by trimming data from unused spectra.  24 bit applies to the whole source and how often the analog source is sampled.

post #158 of 257

24 bit means a higher dynamic range. If I recall correctly, 16 bit has a 96dB dynamic range, meaning if you want to hear it all, the louder parts would damage your hearing. It has everything to do with noise, as any detail that is not within the dynamic range must, by definition, be in the noise floor.

 

Now back on topic, I've listened to the samples and I think I hear better transient detail on sample 1, which I believe is the one sourced from lossless. I listened on my DT 48 through an impedance adapter directly out of my laptop.

 

128kbps does great with tracks that aren't highly detailed, as in there isn't much spectral information to save. It does especially well with tracks that don't make use of high frequencies a lot. Try 128kbps CBR on a highly layered techno mix and tell me you don't hear the difference!

post #159 of 257

24 bit means how many rectangles you fit under the curve of a sinewave.  16 bit have few samples obviously.  Its not just dynamic range to my understanding.


Edited by Anaxilus - 7/16/10 at 6:55pm
post #160 of 257
Quote:
Originally Posted by Anaxilus View Post

24 bit means how many rectangles you fit under the curve of a sinewave.  16 bit have few samples obviously.  Its not just dynamic range to my understanding.


And the biggest change that comes from that is the number of different steps between nothing and as loud as it can go which is more dynamic range.  Its 2^24 = 16777216 vs 2^16 = 65536 so 24 bit holds 16711680 more steps than 16 bit, which translates to a lot more dynamic range.  There is more detail in 24 bit, but I don't know how much of that, if any is audible to actual humans.  It wouldn't surprise me if people with good ears and nice gear could differentiate 16 form 24, but I would be surprised if those same people could tell the difference between 20 and 24.  I think 24 bit is a little bit overboard.

post #161 of 257

Agreed 20 bit is quite nice.  I think I can tell 16bit to 20bit.  20bit to 24bit, I don't think I could. 

post #162 of 257

http://en.wikipedia.org/wiki/Dynamic_range#Audio

 

With Q as the bits per sample, the formula to determine the dynamic range is 6.02*Q. That means that 16 bits per sample has 96.32dB of dynamic range, and 24 bits per sample has 144.48dB of dynamic range.

 

While it sounds like 24 bits means a lot of detail, do bear in mind that it's only 256 times more, and in terms of sound levels and dynamic range, it means 48.16dB more dynamic range.

 

To put this into perspective, 0dB is the threshold of human hearing. Anything below that is inaudible to us. 85dB is the limit where hearing damage begins to occur. That means, we can safely perceive only 85dB of dynamic range. That's already below the dynamic range of 16 bit audio, so whatever detail is contained in there can't even be heard unless you crank the volume to unsafe levels.

 

24 bits per sample has 144.48dB of dynamic range. We can only safely listen to 115dB for 15 seconds. Go figure.

 

Basically, any improvement that you hear from going over 16 bits per sample is either due to unsafe listening levels, or simply due to the placebo effect.

post #163 of 257
Thread Starter 
Quote:
Originally Posted by dsf3g View Post


 


OK, so I went ahead and ripped a few tunes using iTunes and did some testing with The Offspring's "I Want You Bad." I did a double-blind test using Foobar2000 and the ABX plug-in on a 96k file and a 128k file. This was on a Dell Mini 9 netbook using a pair of Sennheiser HD-238 phones. The good news for me was that I was able to distinguish between the two tracks 100% of the time. I did 5 passes and scored 5/5 (3.1% probability of lucky guesses).

 

The bad news for me is that, though I could detect sonic differences between the two tracks, I couldn't tell by listening which of the two was the higher bitrate track. One sounded a bit airier and thinner and the other sounded a bit darker and duller. But I had to really, really pay attention to pick out the differences.

 

The long and the short of it is: I was wrong in thinking I could pick out a 96k track without a frame of reference (a higher bitrate track to compare it to). 96K really is a lot better sounding than I thought.

 

I went ahead and ripped a track in 64k MP3 and 192k AAC (which is what I've got on my iPod). Here, of course, the difference is night and day. On the 64K track the cymbals in the intro simply disappeared. Distinguishing between the 128k MP3 and the 192k AAC was another matter altogether. I didn't run any actual tests, but based upon a quick listen I doubt I'd do much better than chance.

 

So I have to assume that the poorly encoded Led Zeppelin CD I had was encoded at 64K and not 96k.


Just one thing about your test. iTune's MP3 encoder probably sucks. Use LAME. Foobar2000 has an internal GUI for LAME that makes LAME MP3 encoding very easy (just make sure to not use VBR new (or fast mode)).

 

I can tell a difference between 96Kbps and 128Kbps, but not 128Kbps vs >128Kbps 

post #164 of 257
Quote:
Originally Posted by Soaa- View Post

http://en.wikipedia.org/wiki/Dynamic_range#Audio

 

With Q as the bits per sample, the formula to determine the dynamic range is 6.02*Q. That means that 16 bits per sample has 96.32dB of dynamic range, and 24 bits per sample has 144.48dB of dynamic range.

 

While it sounds like 24 bits means a lot of detail, do bear in mind that it's only 256 times more, and in terms of sound levels and dynamic range, it means 48.16dB more dynamic range.

 

To put this into perspective, 0dB is the threshold of human hearing. Anything below that is inaudible to us. 85dB is the limit where hearing damage begins to occur. That means, we can safely perceive only 85dB of dynamic range. That's already below the dynamic range of 16 bit audio, so whatever detail is contained in there can't even be heard unless you crank the volume to unsafe levels.

 

24 bits per sample has 144.48dB of dynamic range. We can only safely listen to 115dB for 15 seconds. Go figure.

 

Basically, any improvement that you hear from going over 16 bits per sample is either due to unsafe listening levels, or simply due to the placebo effect.


Ok, maybe I don't understand something basic here.  Disregarding the whole dynamic range issue for the moment.  The explanation of 24bit sampling as was presented to me by certain Audio companies about 6-7 years ago one of the points of higher bit rates is to increase the resolution of a digital sample and get it closer to an analog curve.  If we imagine a normal analog curve with a peak and a valley a 16 bit digital will sample that section into 16 distinct areas (rectangles) just like approximating the limit of a curve in Calculus.  Everything not sampled by the rectangles under the curve is lost audio data.  By sampling at 24 bits you get 24 rectangles which is 50% more data collection for the same sample section.  Hence you create a smoother sampling with more resolution closer to the original analog signal.  This sampling would apply throughout the entire sample regardless of the volume level or frequency in question right?  I don't understand how the increased data or resolution attained throughout an entire song sampled at 24 bits only exists at a certain dB level?  What am I missing here? 

 

By reading the following along with your previous post:

 

 

"A set of digital audio samples contains data that, when converted into an analog signal, provides the necessary information to reproduce the sound wave. In pulse-code modulation(PCM) sampling, the bit depth will limit quantities such as dynamic range and signal-to-noise ratio. The bit depth will not limit frequency range, which is limited by the sample rate.

By increasing the sampling bit depth, smaller fluctuations of the audio signal can be resolved (also referred to as an increase in dynamic range). The 'rule-of-thumb' relationship between bit depth and dynamic range is, for each 1-bit increase in bit depth, the dynamic range will increase by 6 dB (see Signal-to-noise ratio#Fixed point). 24-bit digital audio has a theoretical maximum dynamic range of 144 dB, compared to 96 dB for 16-bit; however, current digital audio converter technology is limited to dynamic ranges of about 120 dB (20-bit) because of 'real world' limitations in integrated circuit design.[1]"

 

Its is clear that 24bit extends the maximum dynamic range to 144dB.  But that is the maximum.  I don't see anything that says increased resolution does not exists at any level below the maximum of 144dB.  Are you saying at 96dB a 24 bit sample has the same amount of data and resolution as a 16 bit sample?


Edited by Anaxilus - 7/16/10 at 9:49pm
post #165 of 257

 

Quote:
Originally Posted by Anaxilus View Post

24 bit means how many rectangles you fit under the curve of a sinewave.  16 bit have few samples obviously.  Its not just dynamic range to my understanding.


The bit depth tells you the potential dynamic range.  How many rectangles you can stack vertically under the loudest possible sound.  In 16 bit audio you can stick 16 rectangles under the loudest possible sound of 96 dB.  In 24 bit audio you can stick 24 rectangles under the loudest possible sound of 144 dB.  In each case each bit represents about 6 dB.  So the height of a rectangle in 16 bit or 24 bit audio is the same.  No difference in resolution there.  Just a difference in how loud you can go. 

 

The sampling rate tells you how wide each rectangle is.

The bit depth tells you how loud you can go, it does not change how tall each rectangle is.

 

And Nyquist tells you that it doesn't matter how wide each rectangle is as long as the highest frequency you want to record or reproduce is half of the sampling rate.  Having a rectangle narrower than that doesn't gain you anything in theory.

 

Here's a introduction to 24 bit vs 16 bit audio.  Note that they're talking about recording.  Being able to set recording levels and have what you want to record be above the noise floor.  Being able to set levels and have some headroom so when the bass drum gets a LOUD thwack at the end of the song the recording level doesn't go into the red and clip.  Once the recording is done you can optimize the audio to fit in 16 bits.  Since it is already recorded you know exactly what the loudest sample is going to be.  You can optimize for that.  You don't need to reserve any headroom above that.

 

Record at 24 bits.  Deliver to the consumer at 16 bits.

 

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › Double blind test 128Kbps vs lossless? I'll be amazed if you can tell much difference