1. If you are working with 32bit float or 48bit fixed you still have to get back to 24bit with dither or truncation. Logic suggests that the higher the bit rate the higher the quality.

2. SDII may be the format used by more pros than Wav. However, Wav files are not inherently of any less quality that any other format.

3. Dither is ALWAYS bad! Dither is essentially the adding of noise, and who wants to add noise to their mix? Going from say 24bit to 16bit or from 32bit float to 24bit requires the removal of the required number of bits, this is called truncation. Rounding errors during the truncation process can cause audible quantisation errors. Dither (adding noise) can help to mask these errors. Over the last few years dither has been greatly improved with noise shaping where the added noise is moved into a frequency outside the normal hearing range. So in short, dither is essentially a bad thing but is usually less bad than truncation.

Greg

Heres my somewhat confused thoughts on the subject:

1) Float vs Integer
Using float is a bit weird, since float numerics have higher resolution at levels close to zero. If I understand this correctly, that basically means that whatever level the sound is at, you get the same practical resolution, which isn't 32 bit at all. The resolution would rather be 24 bits, the size of a typical mantissa (including the sign bit).

Now, this sounds like a good idea, at least for internal use, but of course, the A/D converter has to work with integers by design, so using float for recording seems strange to me. It makes more sense to record the sound in some integer format, but use 32-bit float for internal calculations.

2) SDII vs Wav
I don't know why one of them should be more "professional", they are pretty similar as to what they can do, AFAIK.

3) Dither is always better than not dithering, when converting to another bit-depth or sample frequency. You would not need it except when you convert, because it's a conversion technique...

howdy...another opinion (well, #2 is a fact i'm pretty sure)...

1)...if you check out the Digi user conference or, say, one of the techier guys' forums at musicplayer.com, or rec.audio.pro archives, and do searches, you certainly should be able to find some pretty interesting and in depth discussions concerning the ups and downs of both fixed and floating point mathematics as they relates to digital audio. i've seen them in the past, and all that i read still left me with no clear preference...

2) SDII and WAV files are the SAME THING! only the header info is different...

3) Dither is neither inherently good nor evil (depending on where your views are towards digital audio in general, that is)...it IS noise, so if applied too often you get a build up, but when used as intended it can improve digitals poor performance concerning low level signals...

Originally posted by proaudio101:
Maybe some of you experts (technically or
1)32 bit Float doesn't do what it says it does (always land at 24 bits), and that 48 bit fixed (Pro Tools) method is better. Whats the reality?

3) Dither is ALWAYS good. When is is good and when is it bad? Why would I need it if I am recording my mulitrack @ 24bits, except to go to the final 16 bit CD.

You need it to linearize the truncation. Otherwise you generate louder distortion than the amount of dither that would have been necessary to linearize. Contrary to common opinion, dither is just as necessary to digital processing as bias is to analog tape. Lazy software developers make you turn it on and off manually rather than detecting and implimenting dither when needed.

Originally posted by Greg Malcangi:
1. If you are working with 32bit float or 48bit fixed you still have to get back to 24bit with dither or truncation. Logic suggests that the higher the bit rate the higher the quality.
Greg

First half is true, second half isn't the logic you might think. 32-bit floating point provides a higher dynamic range (1529 dB since you have the option of exponential arithmetic (24-bit mantissa, 1 sign bit, 7 data bits ). The SQNR (signal to quantization noise ratio) for 32-bit float is fixed at 144dB, regardless of amplitude. 48-bit fixed point provides a fixed dynamic range, and max SQNR of 288dB (SQNR varies depending on signal amplitude - the lower the signal the lower the signal to noise ratio). The big issues in bit depth are with EQ, or similar algorithms, where you can easily develop a noise floor greater than the lowest bit in your audio bit depth, depending on the type of EQ, order, coding etc. Since that noise can accumulate over multiple processes and with multiple signals (such as in a mixing engine/summing bus), bit depth of the mixing engine, and plug-in algorithms are done at higher bit depths to effectively push the noise floor created by that algorithm down well below the final audio bit depth. With fixed point, to handle the above issues with EQ, you generally want double the original bit depth (i.e. double precision) - hence 48-bit (2 24-bit words used together to represent a 48-bit result of a multiply/addition/subtraction,etc). In theory both actually accomplish a similar goal with different mathematical approaches - lower the noise floor of quantization errors by increasing the number of bits used, but floating point offers a fixed SQNR with a higher dynamic range, which is beneficial, since the SQNR of 24-bit audio is 144dB, you never drop below that. With fixed point, you have a lower dynamic range, and the possibility of dropping below the 24-bit SQNR on low signals.

Another issue is when a DAW must dither during the mix/plug-in process, vs. using higher resolution math throughout the mix process and then dither on the final output. Floating point decreases the need for dithering, since it is only needed when converting back to 24-bit integer form. Double precision, fixed point processing of 24-bit words must dither each time a 24-bit word is processed since the single precision bit depth is 24-bits (not 48).

Hmmmm..so you can imagine what's happening with 64 bit plug ins... :roll: