AF016 - Digital Bit Depth Part 2

[@curtpiazza1688]

Great explanation! I like the low-tech approach! 😂

[@ThierryLalinne]

Brilliant! Thanks for all the efforts you put into these great videos. I'm learning a lot.

[@j7ndominica051]

32-bit integer exists. It can be selected in Sound Forge, used as output format on my E-MU ASIO sound card (maybe others). And SoX still works at 32-bit integer, and downconverts output from floating-point effects with clipping.

[@glev7588]

I'm sure you know this, but in 32 bit FP, 1 bit is reserved as a 'sign' bit, 8 to an exponent, and 23 to the mantissa (number), yielding a range of (2^1) * (2^8) * (2^23), or 4,294,967,296 (−2,147,483,648 to 2,147,483,648). Hope you find that interesting! Thanks for the videos!

[@LoudnessWar]

People are LEARNING! YES. My diabolical master plan is working! MWAAHAHAHAhahahaaa

[@officialrecuest]

Hi, I want to start off by thanking you soooo much for making these videos. The information and explanations are fantastic, I really hope you keep making more :D I have a question for this video. I assume a sample with a 100% value means the speakers are pushed in/out as far as possible when going through a D/A converter. When a sample's value is at 100% and the channel/master's at 50% (-6db gain right?).

[@LoudnessWar]

Thanks a lot. Glad it's helpful.

[@officialrecuest]

I can' t amplify the value of the sample without clipping and getting a distorted sound (unless I use dynamic compression which still wouldn't affect the sample at 100% xD). So what is happening when I turn the gain up to 100%? I suppose that turning the volume up on some studio monitors is in the analogue domain. But if we do so in a DAW or something, does it mean that the 100% sample is "de-amplified" (by dividing its value) when the gain is turned down. How exactly does all of this work? :S

[@syncsound]

Love the series! One thing I'm still hung up on (and I do this for a living!) is the bit depth analogy: in the video, you use the ruler, and describe 24 bit as having smaller steps, thus allowing it to quantize more accurately. But according to the math (which is over my head), each bit yields 6.02 dB in dynamic range. If each bit has a fixed dynamic range yield, how can 24 bit offer "smaller steps"?

[@LoudnessWar]

That's a great question. The idea is that adding and subtracting dB really means *multiplying* and *dividing* the samples, and that the added dynamic range improves how *quiet* you can get before it all rounds off to total silence. For example, let's say we have a sample value of 10. Here's an analogy - it's very sloppy but I hope it gets the idea across. Low bit depth analogy: no decimal places 10 -> 5 -> 2 (round down) -> 1 -> 0.... and then trying to divide always gives 0; we've lost the signal completely. We're out of dynamic range after 4 cuts in half, so it's 4*6ish = about 24dB usable dynamic range from Full Scale to the dropoff point. (Did I say the analogy was very sloppy?) Higher bit depth analogy: 1 decimal place 10 -> 5 -> 2.5 -> 1.2 -> 0.6 -> 0.3 -> 0.1 -> 0... where we lose the signal. Having an extra decimal place (kind of like having a few more bits) gives us 7 halvings before we're stuck at 0, so it's 7*6ish = about 42dB. Very high bit depth analogy: 4 decimal places 10 -> 5 -> 2.5 -> 1.25 -> 0.625 -> 0.3125 -> 0.1562 (round down) -> lots more times before it hits 0 = very large (more than 100dB) dynamic range I hope that helps; the Dynamic Range By Domain videos touch on this as well, if you want an audio example.

[@syncsound]

That does help, thank you! I think I got hung up on the erroneous idea that each change in amplitude would occur in coarse 6db increments, when each additional bit simply offers *up* to 6db of gain. The decimal analogy really helped. Great series!