xiphmont’s ‘There is no point to distributing music in 24 bit/192 kHz’, Article: “24/192 Music Downloads are Very Silly Indeed” Options

[NullC]

Of interest, but certainly not news to anyone here is Monty's educational piece regarding the lack of value of 24/192 as a distribution format:

http://people.xiph.org/~xiphmont/demo/neil-young.html

Some of the commentary on the internet has been not especially well informed, for example there are some crazy claims at https://news.ycombinator.com/item?id=3668310. If you're in a "someone is wrong on the internet!" correcting mood, you might want to go leave some comments in furtherance of the collective intelligence of mankind.

Cheers

[pdq]

I would hardly call float32 a danger zone. Even when the signal is greater than 0.5 of full scale (which is a small percentage of the time) float32 has a full 24 bits of resolution. When the signal is down at 2^-24 you still have 24 bits, that is 2^-48 of full scale.

When you perform repeated arithmetic operations on the data, if the rounding is uncorrelated (which it should be) then the increase in digitization error only increases as the square root of the number of operations. One thousand operations increases the error by 5 bits.

[icstm]

I would hardly call float32 a danger zone. Even when the signal is greater than 0.5 of full scale (which is a small percentage of the time) float32 has a full 24 bits of resolution. When the signal is down at 2^-24 you still have 24 bits, that is 2^-48 of full scale.

When you perform repeated arithmetic operations on the data, if the rounding is uncorrelated (which it should be) then the increase in digitization error only increases as the square root of the number of operations. One thousand operations increases the error by 5 bits.

that sounds about right and even then that is worse case, as you could simplfy the operations before performing them in the first place thus limiting the places where rounding occurs.