A question for the experts!
I record at 24 bit and mix to 32 bit -then do whatever edits are needed in Sound Forge 6.0. I then do bit depth reduction in Sound Forge to 16 bit. Sound Forge offers 5 types of dither - Rectangular 1 bit peak to peak, Rectangular 2 bits peak to peak, triangular 2 bits peak to peak, highpass triangular and gaussian 2 bits rms to rms. I’ve been using highpass triangular and it seems to work fine but I was wondering what is considered best. Any insight would be appreciated!
Use whatever sounds best. The reason they give you differnt dithers is so that you have options. Certain dithers can sound better depending on the material. Reading the help that comes with Sound Forge (F1 is your friend), this is what they say:
<!–QuoteBegin>
| Quote |
| Rectangular Eliminates distortion caused by conversion to a lower bit depth, but the noise level is dependent on the signal. Triangular Eliminates distortion caused by conversion to a lower bit depth and eliminates noise floor modulation by producing a slightly higher noise level. Highpass Triangular Eliminates distortion caused by conversion to a lower bit depth and eliminates noise floor modulation by producing a slightly higher noise level. Noise is shifted to higher frequencies than standard triangular dithering. Gaussian Does not perform as well as rectangular or triangular dithering, but may be suitable for some material. In general, Highpass Triangular with noise shaping produces the most favorable results. |
| Quote (BobbyS @ Mar. 11 2005,16:36) |
| A question for the experts! |
I'm the exact opposite of an expert on dithering, hence my question: What is dithering/noise shaping and why do I need it?
Thanks
In popular terms, dithering gets rid of the effects of quantization effects experienced in digital recordings. When dealing with a 16 bit recording, the signal may take one of 65.536 unique, discrete levels, the signal is quantized. This may in some cases lead to a ‘cold’, ‘sterile’, or even unnatural-sounding result. The more times the sound is processed, the worse the effect.
Dithering adds a tiny, random signal (usually only one bit - hence the ‘bit’-setting in n-Track dithering options) to the original signal. While this theoretically adds to the noise figure, in practice, it evens out the negative effects of quantization.
The noise shaping is a way of ‘weighing’ the dithering, so that small signal levels are dithered differently from large signal levels. This makes the signals sound more ‘open’ or 'natural’
For fun, try making a ten second recording of silence, then render the track using a dithering of, say, 8 bit. -> Pure white noise… 
regards, Nils
What Nils said is basically correct, but doesn’t cover both reasons (reducing q-error versus masking it).
When you’re reducing bit depth, you’re increasing the quantization error amount. I.e., if you’re going from 24 bit to 16 bits, instead of 256 tiny steps between two nearby values, there’s one step that’s 256 times as big. Lopping off those low order 8 bits (by rounding) increases the quantization error for each sample. Dithering is adding a random number before the rounding process, and causes the overall average introduced error to be half of what it would have been (for 1-bit dithering).
If you recorded the signal in 16-bit mode, did NO processing to it at all, then you’re not increasing the quantization error at the end, so you don’t need dithering for that purpose. However, there still is that quantization noise down there at the bottom, and it’s a very nasty sounding kind of noise. Therefore, we often add lovely sounding white (or nearly white) noise to mask the harsh q-noise.
Folks usually don’t distinguish between the two reasons, but for the first reason you’d only ever add 1 bit, and for the second reason you might add more. However, we almost always process our signals and that causes the signal to be more than 16 bits wide, so we always dither at least one bit as the very last step of the mastering process, or whenver we’re forced to reduce bit depth. The signals I record usually have enough natural noise in them so that I don’t need to add more to cover up the q-noise. But processing (FX & EQ) can often exacerbate q-noise, so maybe we DO want to dither more bits.
This is why it’s a matter of taste and “use what sounds good” rather than an exact science. You should dither by at least one bit by default, and maybe 2. I set this in the “options” menu for dithering on playback, so it’s set that way by default when I mixdown.
FYI, the “bit” setting in n-Track is NOT a number of bits, it’s an integer value, left justified in the target format. Eg, 1 gets you 1 bit, 2 gets you 2, but 4 gets you 3 and 8 gets you 4, etc. Also, last I tested it, there were some anomalies; differences between mono and stereo mode that I wouldn’t expect, and the resulting number of dithered bits was half or twice what you’d expect in one mode or the other.
I should retest with V4 to see what it does now and submit a bug report if it’s still true.
Nils, I hadn’t heard noise shaping described that way. Instead, I’ve heard it as being frequency-shifted so that more energy is in the higher frequencies than white noise, thus making it less audible. In any case, there are a number of different “noise shaping” algorithms; I sure don’t know which is best.
Wow…There’s some very knowlegable people using n-track. I did’t really follow what you said though.
When I mix down on n-track and at the same time choose to convert to 16 bit (from 24) it has a check box that says “use dither”. Up to now I’ve always ignored that and been happy with results. Should I be checking the dither box?
| Quote (true north @ Mar. 14 2005,14:13) |
| Wow..There's some very knowlegable people using n-track. I did't really follow what you said though. When I mix down on n-track and at the same time choose to convert to 16 bit (from 24) it has a check box that says "use dither". Up to now I've always ignored that and been happy with results. Should I be checking the dither box? |
As long as you are happy with the results, keep on keeping on! There has been much discussion and hair-pulling over dither. Personally, I have found that the ears are the ultimate judge. Most all programs apply dither in some form or fashion without you even realizing it's there. Otherwise, the recorded audio would sound like poo once played back. Dither at conversion from 24 to 16 bit? Use your ears. If it sounds good, it is good.
TG
I would highly suggest you read the Ozone Dithering Guide. It is specific to iZtope’s Ozone product in some aspects, but the basic theory they teach you applies to all dither in digital audio. It is very good in that it shows you pictures and graphs so you can visualize what in the world we are talking about much better. A must read. Really, it is excellent. Go read it… now! 
Do what Bubba says there…it’s been a while since I looked that over. It is indeed a great resource. Also, note the STEPS they indicate when applying dither. It is super easy to undo what you just dithered if you don’t pay attention when using a dither plugin like Ozone.
Thanks for bringing that back to the table Bubba. It’s good to refresh the ol’ gray matter every once in a while.
TG
I agree, the Ozone guide is a must read… Have you read it yet?
teej
| Quote (Bubbagump @ Mar. 14 2005,14:36) |
| I would highly suggest you read the Ozone Dithering Guide. ...It is very good in that it shows you pictures and graphs so you can visualize what in the world we are talking about much better. A must read. Really, it is excellent. Go read it... now!
|
Wow, that *is* a nice doc...thanks, Bubba! Uggg...Pictures and graphs GOOD!
Tony