[Sam Spade]

.
a. double blind testing isnt a good testing method

Please explain why

 

[gregorio]

Please explain why

Not sure you’re going to get a response to that, more than 14 years after he posted it!

G

 

[71 dB]

Please explain why

You quoted a 14 years old post. Don't hold your breath while waiting for a response to that from the original poster...

It is funny how the subjectivists whine about the "problems" of double blind listening tests while being happy with the serious confirmation bias/placebo-effect-type problems of non-blind listening tests.

 

[Sam Spade]

You quoted a 14 years old post. Don't hold your breath while waiting for a response to that from the original poster...

It is funny how the subjectivists whine about the "problems" of double blind listening tests while being happy with the serious confirmation bias/placebo-effect-type problems of non-blind listening tests.

Yeah I just Figured something so stupid had to have an entertaining explanation

 

[71 dB]

Yeah I just Figured something so stupid had to have an entertaining explanation

I have to say I know a lot of entertainment better than explanations as to why double blind listening tests aren't good.

 

[texx2818]

My point in all this is WHY compromise anything at all?

Simply stay in 24 bit.

He doesn't know. Conjecture. Data and "can't hear it"

 

[gregorio]

He doesn't know. Conjecture. Data and "can't hear it"

I do know and what conjecture?

You got “data” and inaudible right though!

G

 

[Ryokan]

Could someone explain 0dBFS and why it's the highest level and anything above will clip.

 

[danadam]

Could someone explain 0dBFS and why it's the highest level and anything above will clip.

A decibel is a ratio between 2 values on a logarithmic scale. It can be used either to express a relative change between any 2 values, we then use "dB" unit without any suffix, or it can be used to express an absolute value. In that second case one of the values in the ratio, the denominator, is fixed to some reference value and then we use "dB" unit with some suffix, e.g. dBSPL for absolute values of sound pressure level or dBFS for absolute values of digital level.

For SPL the reference value is 20 uPa. Then

• 0 dBSPL means absolute value of pressure equal to 20 uPa (no change from the reference),
• +6 dBSPL means absolute value of pressure equal to 40 uPa (twice the reference) and
• -6 dBSPL means absolute value of pressure equal to 10 uPa (half the reference).

For FS in case of audio encoded using integer types, the reference value is full scale, i.e. the maximum that an integer sample can hold. Then

• 0 dBFS means absolute value of digital level equal to maximum possible and
• positive dBSPL is not possible to represent in the integer type.

For FS In case of audio encoded using float types, the reference value is 1.0. Then

• 0 dBFS means absolute value of digital level equal to 1.0,
• +6 dBFS means absolute value of digital level equal to 2.0 and
• -6 dBFS means absolute value of digital level equal to 0.5.

1.0 is still considered to be full scale, i.e. if you convert it to integer type then value 1.0 becomes the maximum value that this integer type can hold. If you have your audio encoded in float type and some of the values are bigger than 1.0 (or in other words are above 0 dBFS) then during conversion to an integer type they will be clipped to 0 dBFS.

 

[Ryokan]

+6 dBSPL means absolute value of pressure equal to 40 uPa (twice the reference)

Why is 6dbspl twice reference, what would 8dbspl equal?

 

[gregorio]

Could someone explain 0dBFS and why it's the highest level and anything above will clip.

Danadam’s explanation is correct. In even simpler terms though: 0dBFS represents all the bits set to “1”. IE. All 16 bits or all 24 bits. Obviously there can be no value higher than all bits set to “1”. The obvious exception is float format, a 32bit float has 24 bits for the mantissa and 8bits for the exponent. The exponent allows values to be represented far in excess of 0dBFS (all the mantissa bits set to “1”) but as danadam stated and I did previously, any such value has to be reduced to below 0dBFS on conversion.

Why is 6dbspl twice reference, what would 8dbspl equal?

The decibel scale is logarithmic. When representing voltage, sound pressure levels, the digital scale and others, +6 is double, +20 is ten times and every additional +20 is an additional factor of ten (so +60dB is 1,000 times). Note that some dB scales (such as those representing power, the dB scale for watts for example) use use log base 10 rather than log base 20, so +3dBW is double and +10dBW is a factor of ten times.

G

 

[VNandor]

Why is 6dbspl twice reference, what would 8dbspl equal?

It comes from how people decided to define decibels. In the case of of sound pressure levels it's defined as dBSPL = 20*log(p/p0) where p0 is the reference value (20uPa) and p is the rms sound pressure you want to express in dBSPL.

The above equation can be "solved" for p instead dBSPL. It would be p= p0*10^(dBSPL/20). If you plug in 6dBSPL into this equation, you would get 39.9uPascal which is almost twice the reference. So this is why 6dBSPL is twice the reference. Essentially 10^(6/20) is close to 2 so that's why people say "6dB is twice as much". You can find out how much Pascal 8dBSPL is by plugging in that into the equation instead. That would be around 50.2uPascal.

 

[71 dB]

Danadam’s explanation is correct. In even simpler terms though: 0dBFS represents all the bits set to “1”. IE. All 16 bits or all 24 bits. Obviously there can be no value higher than all bits set to “1”.

The first bit is sign bit: 0 = +, 1 = -. So, 0111111111111111 would be "max value" in 16 bit. For example in 4 bit Two's complement +7 = 0111 (max value) while -7 is calculated by flipping all bits and adding 1 to the result: 1000 + 1 = 1001.

.......................-8 = 1000
+7 = 0111 => -7 = 1000 + 1 = 1001
+6 = 0110 => -6 = 1001 + 1 = 1010
+5 = 0101 => -5 =1010 + 1 = 1011
+4 = 0100 => -4 = 1011 + 1 = 1100
+3 = 0011 => -3 = 1100 + 1 = 1101
+2 = 0010 => -2 = 1101 + 1 = 1110
+1 = 0001 => -1 = 1110 + 1 = 1111
0 = 0000

 

[71 dB]

Why is 6dbspl twice reference, what would 8dbspl equal?

10^(6/20) ≈ 1.995262315 ≈ 2
10^(8/20) ≈ 2.511886431 ≈ 2.5

To answer your question, about 50 𝝁Pa (2.5 * 20 𝝁Pa).

 

[gregorio]

The first bit is sign bit: 0 = +, 1 = -.

Yes, I was trying to keep it simple so it was easy to conceptualise. The same is also true with the mantissa of a float format.

@Ryokan there’s some interesting things about the decibel scale:
1. It was invented in the 1920’s to replace the previous scale used to measure signal loss in Miles of Standard telephone Cable (MSC).
2. The decibel was named in honour of Alexander Graham Bell, a decibel is a tenth of a “Bel”.
3. The sengpielaudio website has a number of useful decibel calculators and converters. Here’s the page for converting dB values into factor (ratio) values. Your 8dB comes out at just over 2.5 times.
4. As the decibel scale is logarithmic, we can’t just add different decibel values together. This makes dB calculations very unintuitive and misunderstood by most audiophiles. Let’s take 2 examples:

Example A:
Let’s say we have a trumpet playing loudly, which measures 100dBSPL (at say 5m away). Let’s say we then add 20 violins playing loudly, each of which individually is producing 80dBSPL (also at say 5m away). What’s the dBSPL level now? Obviously it’s not 100 + (20 x 80) but maybe we can use our knowledge of factors/ratios to figure it out? 80dB is 20dB less than 100dB and therefore 10 times less, so 10 violins would be the same (100dB) level and 20 violins would be 10 times (20dB) more, so our answer is a bit above 120 dBSPL. Does that sound reasonable/logical? That’s still completely wrong, the actual answer of a 100 dB trumpet plus 20 violins each at 80 dB is 100.69 dBSPL!

Example B:
Let’s take our noise floor/dither discussion. Let’s say we have a very dynamic symphony recording, with a 60dB dynamic range and play it at a peak level of 100dBSPL. The noise floor will therefore be at 40dBSPL. Let’s add to that the digital noise floor, using traditional/triangular (not noise-shaped) dither (at typically around -90dBFS) and therefore at our playback level we would have 10dBSPL (100 - 90) of dither noise. So what is 40dB of noise plus 10dB of (uncorrelated) noise? It’s actually 40.0043 dBSPL of noise! Baring in mind that 0.7dB is a just noticeable difference (JND), then our 0.0043dB difference is way below audibility! Although expressed in audiophile terminology: The blacks are no longer black and anyone with decent ears and audiophile equipment would notice, even your wife! Lol.

G