32 bit floating wav vs 32 bit

[mahimike]

OK- I am probably over-analyzing: the sample workflow for vinyl recording says to use wav 32 bit float.
1) The initial export choice is 32 bit float PCM
2) other options include: a) signed 32 bit PCM & b) 32 bit float

Which should I be using? I have tried to find out the difference but only get to discussions about the difference between 16/32 bit.

[steve]

During "production" (while you're working on the project), keep to 32-bit float as far as possible. Audacity works internally with 32-bit float, which is a ridiculously high quality format that is ideal for audio production. Working with 32-bit float throughout avoids unnecessary conversions that will inevitably have some quality losses (though not necessarily noticeable losses).

For your final export, go with 16-bit. It's plenty high enough quality, and it is compatible with everything.

[mahimike]

Steve- Thanks for the fast response. I understand the need to stay in 32bit. My question is that there are 3 different 32 bit choices for WAV files when I export. Are they all the same? Is one better than the other? Sorry, my post was not more clear.

Mike

[DVDdoug]

Like Steve says, floating point is better/easier for DSP (digital signal processing) so Audacity (and most similar software) uses floating-point.

Another advantage is that there is virtually no upper (or lower) limit with floating point. You can't go over 0dB when recording or playing, but you can do processing that (temporarily) pushes the peaks over 0dB (such as boosting the bass) and as long as you reduce the volume before exporting, the wave shape will be preserved and the peaks won't be clipped/distorted.

All integer formats are limited to 0dB. If you try to go over 0dB, you'll get clipping. With more bits you get more resolution and more range on the "quiet" side, but it doesn't go higher/louder.

32-bit integer is very rare. I don't know of any 32-bit analog-to-digital converters, and I've only heard of one 32-bit digital-to-analog converters. And, virtually all analog-to-digital and digital-to-analog converters are integer (and therefore hard-limited to 0dB) . So if you record in floating-point, the software/drivers are making a conversion. The conversion between 16-bit or 24-bit to 32-bit floating-point, and back is lossless.

Pros generally record at 24-bit/96kHz, but of course their software uses 32-bit or 64-bit floating point "internally". As you may know CDs are 16-bit/44.1kHz. The guys who do scientific-blind ABX tests have pretty-much demonstrated that you can't hear any difference between a "high resolution" original and a copy downsampled to "CD quality".

[steve]

My question is that there are 3 different 32 bit choices for WAV files when I export. Are they all the same? Is one better than the other?

All 32-bit formats provide excellent quality (virtually "perfect").
32-bit float is the one that Audacity uses internally, and it is the only one that supports signals over 0 dB. Unless you have a good reason to do otherwise, follow the advice in my previous post.

[mahimike]

So 32 bit float is the way to go. So, as an academic question, what is the difference between
32 bit float PCM and 32 bit float? These are two choices in the export options.

[kozikowski]

So 32 bit float is the way to go.

To fold this back to the original question, I'm not sure saving your vinyl archive in 32-float is a good idea. The stand-alone format is not well supported and you could easily find an app that has no idea what to do with it. This is in addition to taking up massively more storage than necessary and being overkill compared to the original vinyl work.

44100, 16-bit Stereo is perfectly supported by almost all apps on all three major computer platforms. If leading edge adventure appeals to you, use 48000 the video standard instead of 44100.

Koz

[steve]

as an academic question, what is the difference between
32 bit float PCM and 32 bit float?

None (assuming that both are PCM), but I suspect that you typed your question wrong.

"32-bit float" refers to how the numeric value of each sample is represented. In this case, the sample values are represented as "32-bit floating point" (binary) numbers. This numeric format is also known as "Single-precision floating-point format" and is defined by the IEEE 754 standard: https://en.wikipedia.org/wiki/Single-pr ... :_binary32
Floating point numbers can represent fractional values.
"Valid" sample values are generally considered to be values below 0 dB.
In 32-bit float, the range of +1 to -1 is used to represent values below 0 dB.
32-bit float is able to represent numbers much greater than 0 dB.

"32-bit signed" is another way to represent numbers with 32 binary digits. In this case, the number format is 32-bit "signed integer".
In this format, the numeric range is from 0000 0000 0000 0000 0000 0000 0000 0000 to 1111 1111 1111 1111 1111 1111 1111 1111.
Negative numbers are represented as "2's compliment" https://en.wikipedia.org/wiki/Two%27s_complement
so the largest positive number is 0111 1111 1111 1111 1111 1111 1111 1111 = 2147483647 (decimal)
and the most negative number is 1000 0000 0000 0000 0000 0000 0000 0000 = -2147483648 (decimal)
The sample values are "normalized" to a range of +1 to -1 by dividing the numeric value by 2147483648 (so the maximum positive value is actually a tiny fraction less than 1).
It follows that sample values beyond 0 dB cannot be represented when using integer format.