Any type of conversion process will alter the sound quality some. If done properly it is usually on the minor side. The biggest negative results come from a drop in Sample or Bit rate.

One analogy would be the digital info "binary code" (01100101) is like a book that represents the sound. With more samples per second and longer bit words (24 vs 16) you get a bigger book with more pages (details), better representation. If you take away some of those pages from the long book the details become fuzzy (sometimes with more noticeable effect, sometimes with lessor an effect). If you take a shorter book and add pages it might not increase much (cannot add what is not there to start) but it will generally not hurt the story.

Most computer editors will do conversions upon import. BUT the dithering or resampling process is better with some Software or Hardware than others. Apogee and TC electronics have some nice selective dithering processes that help maintain the sound quality (during Bit Rate Conversion). When changing the sampling rate drastically it might be resampled. This means any sampling factors such as oversampling rates will effect the final result (specific to what is doing the conversion).

It is best to avoid but if it is unavoidable then just try to use the best Software/Hardware available. Being that it is a digital process all methods will differ in quality and effectiveness.

With samples it is hard to always have the right digital format. But with recordings I usually stick with standard sampling rates if the end product is to CD (16bit-44.1k). I record commonly in 24bit-44.1k with CD projects to avoid sample rate down conversion. The bit rate dither down to 16bit for the final CD master is usually done with Apogee UV22 converters (one with selective dithering). Without the selective bit rate dithering available I would probably start with 16bit.

FYI - for those who are not familiar with selective dithering. It is suppose to select the empty (waste) bits to remove first. Those bits that contain silence and noise, maintaining the bulk of the signal best as possible. Normal dithering just chops 8 bits from the word (24->16) indiscriminately (sometimes taking the good and leaving the junk).

xaivious wrote

FYI - for those who are not familiar with selective dithering. It is suppose to select the empty (waste) bits to remove first. Those bits that contain silence and noise, maintaining the bulk of the signal best as possible. Normal dithering just chops 8 bits from the word (24->16) indiscriminately (sometimes taking the good and leaving the junk).

I think you're a little off here. Dithering is the process of adding noise to a digital signal trading distortion for noise. This to me would be normal dithering.

What selective dithering is I don't know, but your explanation does not go well in line with the math - there are no such things as empty bits. A bit is either 0 or 1 (as represented by a computer, where a zero is just a representation and contains equally as much information as a one) and together the bits makes a word which more or less accurately describes an analog signal. This accuracy is called resolution. Low resolution digital audio sounds a lot like distortion.

Read a very descriptive article here:

http://

Moderators: it seems like the digital domain can be very confusing. Myself (as an electrical engineer who have worked with analog and digital electronics including AD/DA converters) also sometimes have a hard time understanding the process of sampling. I suggest that you create a sticky post explaining the fundamental things like sampling, sample rate, bit depth, sample rate and bit depth conversion.

Also I would love to know what selective dithering really is.[/url]

phulden wrote:
I think you're a little off here. Dithering is the process of adding noise to a digital signal trading distortion for noise. This to me would be normal dithering.

What selective dithering is I don't know, but your explanation does not go well in line with the math - there are no such things as empty bits.

When I posted this I was wondering if someone would pull the literal term of "dithering" out of the box. Thanks for not trying to be insulting or egoistic about it, like some people on here.

I understand the true digital process of dithering. I am a professional recording engineer and have studied digital recording from nyquist theory and beyond.

However, many companies call the process of selectively dropping bit rate, "selective dithering". I worked in pro audio sales years back and it was from the TC and Apogee folk that I heard it used to describe the methods. I cannot say I agree with the term since it has other definitions and is confusing. And they may have changed it since, but it is or was there never the less. Remember some words mean more than one thing and are not always easily referenced in the dictionary (especially with science and engineering). Isn't the use of english language wonderful. So you need to think of the context of the word here. As far as "selective dithering", it has been explained to me just as I put it.

I was not saying that the bits were empty as in nothing there. I was stating the waste bits, or empty of important content. Think of a signal, a kick drum miked up and hit on the 4. Between those hits the level on the meter drops to pretty much nothing, but true silence rarely exists. There is always some sort of noise on the floor. The sampling process usually samples this noise. In fact when a sample falls below a certain level, traditional "dithering" adds noise to fill in the gaps and prevent distortion. The concept as I have been told about the selective bit rate drop (selective dithering) is those such bits, representing noise or other waste, are removed first. Perhaps the term (selective dithering) arose from the fact that it removes any added dithered noise from the digital process first. I am not the engineer behind it so I cannot tell you how it works exactly or the background, I can only provide this vague description.

Digital sampling (recording) like advanced computer programming is very complex. Think of the sampled waveform, a sample is sustained till the next sample occurs producing a rigid (step like) representation. Even the word bits of this rigid (step like) wave are not representing the sound but are used later to reform the fluid analogous waveform using complex filtering and what not.

Then once we all have it somewhat straight in our heads the next generation of methods come along. Sony has for a while now had a new process called Direct Digital Streaming (DDS) that is supposedly more effective than word bits. It is where the binary code is communicated in a stream rather than with pulse code modulation (PCM). One could spend a lifetime becoming an expert (and keeping up) with digital recording theory.

Thanks for taking your time, James. It seems to me that the terms are not really well defined as you point out.

Do I get you correct when "selective dithering" has to do with sample rate conversion (i.e. going from 48k to 44.1k) rather than bit depth decrease (24 bit -> 16 bit)? In that case I see what you mean with "empty bits".

Groff, sorry that we got off topic from your original question. I think I read here somewhere that in case you need to do sample rate conversion, i.e. record at a higher sample rate than the the end production play back medium, you should try to go for doubling the frequency. That is for a CD production record at 88.2 kHz (if supported) and for video at 96 kHz. In this case you can gain some quality provided that you do the sample rate conversion as the last step (after mastering?). Not that I ever think I'd be able to hear that myself but at least the harddisk manufacturers would be happy because they'll have to double their sales.

My personal opinion is that (amateur as I am) my gear and my limited skills will probably be much worse a source of bad sound quality than a sample rate conversion. So I'll just stick to recording and working at 48kHz/24 bit and make the conversion as the last step, because that works for me and the files will not end eating all of my harddisks. Well, I know HDs and DVDs are cheap nowadays but I'd rather spend that money on other gear instead.

Comments James?

Sincerely

Pål

Read a very descriptive article here:

http://

Thanks for that.

Moderators: it seems like the digital domain can be very confusing. Myself (as an electrical engineer who have worked with analog and digital electronics including AD/DA converters) also sometimes have a hard time understanding the process of sampling. I suggest that you create a sticky post explaining the fundamental things like sampling, sample rate, bit depth, sample rate and bit depth conversion

I agree and vote for. I saw too many arguing about it in this forum, too many different opinions and explanations of the same math but without final conclusions especially for the practical use.

Normal dithering just chops 8 bits from the word (24->16) indiscriminately (sometimes taking the good and leaving the junk).

Mjones4th wrote (topic: «24 bit to 16 bit: what's lost?»)

The quietest parts of a digital recording (like the very end of a long reverb tail) use bits 1-8. The middle of the line signal fall somewhere between 9-16. The loud stuff is up there at 17-24 bits.

That is the effect of bit-depth truncation.

Now I'm indiscriminately :? some more.

Still don't know the best way to convert 16/44.1 to 24/96. For the 24/96 to 16/44.1 I'm using Apogee Rosetta 200

The quietest parts of a digital recording (like the very end of a long reverb tail) use bits 1-8. The middle of the line signal fall somewhere between 9-16. The loud stuff is up there at 17-24 bits

Well... that's a way to put it, it all depends on how your gain knob is turned. But, yes, in essence it is correct. Louder and louder audio will affect (what are called) the most significant bits of a digital word.

Still don't know the best way to convert 16/44.1 to 24/96. For the 24/96 to 16/44.1 I'm using Apogee Rosetta 200

Well to me it seems like upsampling should not alter sound quality at all because you really don't remove any information from the sound data. (However some interpolation needs to be done to go from 44.1 to 96 so theoretically the sound would be slightly altered, no idea if this can be heard). What the Rosetta converter really does I don't know, but I would equally well rely on software processing here (for example Adobe Audition or Wavelab should do the job fine).

Increasing bit depth will not alter quality since nothing is removed and nothing is added, just the precision guarantee is greater (it's kind of like telling you that your checkbook balance is $100.00000 instead of $100.00).

However some interpolation needs to be done to go from 44.1 to 96 so theoretically the sound would be slightly altered, no idea if this can be heard...

Increasing bit depth will not alter quality since nothing is removed and nothing is added

I was thinking the same, but that is just my simple logical approach. Those things are far more complicated and not so obvious, so I'm not sure.

I am going to try to post a little more on this subject tomorrow if time permits, but if anyone here wants to read up on the DSP behind sample rate conversion, I would pick up a copy of Richard Lyons' _Understanding Digital Signal Processing_ (get the second edition). This is a book written by an engineer for other engineers. I am a little biased since I helped review the new edition :) , but there is a lot of misinformation floating around the net (and in product descriptions), and this is a really good book.

mpd wrote: I am going to try to post a little more on this subject tomorrow if time permits.

Any advice would be of immense help. Thanks.

phulden wrote:
Do I get you correct when "selective dithering" has to do with sample rate conversion (i.e. going from 48k to 44.1k) rather than bit depth decrease (24 bit -> 16 bit)? In that case I see what you mean with "empty bits"

Comments James?

As far as I have been told the process sometimes known as "selective dithering" found in products like the (Apogee UV22) or the (TC Electronics Finalizer 96k) has only to do with the bit rate conversion (24bit>16bit). So it does not apply to the sample rate conversion. The empty (waste) bits was referring to lessor important parts of the bit word ("empty" of important content, not of data).

I have never studied much into the sample rate conversion process in detail or the latest things available. Apogee is usually some of the best. So Groff, if you are using the Apogee Rosetta you should be in good shape. As far as I know, sample rate conversion is still just a resampling process. So the better the equipment, hardware or software (higher oversampling rates, etc.), the better the result. And to anyone needing to do conversions realize cheaper equipment could compromise overall quality.

Mjones4th is correct about the bite rate drop not being exact and varying. I was just being really general as to not get into bit word formation saying (8 from 24). If you get technical, of course, it is more complex. Technically on a digital meter you only aquire 24 bits when the meter is full. There are all so many contributing factors.

Commenting on why doubling the sample rate is so much better (like 88.2khz/96khz). In theory recording at twice your current standard sampling rates (44.1-48khz) allows for a more full 20+khz harmonic structure to be recorded (sampled) digitally.

This comes from Nyquist theory. When sampling, there is a point roughly halfway between the sampling frequency (example 44.1khz) that the signal will distort (producing "digital distortion"). So a special Band Pass filter (sometimes called the "nyquist filter") cuts all frequencies to zero passing the half way point (example 44.1khz - ->22.05khz). Though technically the sound spectrum ends around 20-22k, higher harmonics past 22.05k do affect the tonal structure of a complex waveform (like a guitar note) . This upper harmonic effect can be very present up to some 40k (but likely even more). So when you record even at 48k you are missing the full harmonic tone, removed by the nyquist filter (at 24k). When you move to 96k (even 192k), you are doubling or tripling the threshold of the nyquist filter, allowing for a more full sampled waveform. So you are getting a more full and accurate representation of the sound, always better.

However, when your end product is a CD 16-bit/44.1k (or DVD-Video 24-bit/48k) you are going to lose that detailed 96k sample either way.

So, is it worth it? Technically you might be able to preserve a little of that really good sampled 96k waveform with high-quality conversion. But with the extra time and needed equipment I personally do not think it is worth the trouble to go 96k right now. Maybe to archive for a remix in a few years to DVD-Audio (24-bit/96k), that is different. Then (if you got the resources) it is definitely worth going higher sampling rates and coverting down now for CD/DVD production. Since DVD-Video is 24-bit I definitely think it is smart to record with that bit resolution either way at current. Eventually the sampling rates will permanently shift, but not yet.

Please,
could I ask for a favor. Please stop using the term "selective dithering". It is not a standard term used in the literature, it might be something used by only one company as far as I know. The word and the description, in my mind, only confuses things.

Secondly, please don´t use the term "bit rate" either. It is also bound to make for confusion. (Bit rate is used in a completely different meaning when you are talking about networks or radio communication).

I believe better terms to use are:
sample rate -- measured in Hertz (or kiloHertz when that applies) for how often samples are taken of the input signal.

My English beeing a foreign language I have no really one-step name for the number of bits in a conversion as a concept, perhaps someone could help here. Sample depth is the word I tend to favor.

------------
Now, the questions was, how to change the sample rate in as good way as possible. This is generally called sample rate conversion or SRC. This will inevitably influence or change the sound somewhat, and that is why different ways of doing it might make a change. So try the different audio programs you have access to and, and listen for yourself if you can hear a difference and in that case which one you preferr. It is a good idea to go up to 24 bit if the rest of your project is at that rate.

If you have access to WaveLab you might try that. I generally tend to use my DAW which is Samplitude, as it is good in this area.

Gunnar.

ghellquist wrote: Please,
could I ask for a favor. Please stop using the term "selective dithering". It is not a standard term used in the literature, it might be something used by only one company as far as I know. The word and the description, in my mind, only confuses things.

Secondly, please don´t use the term "bit rate" either. It is also bound to make for confusion. (Bit rate is used in a completely different meaning when you are talking about networks or radio communication).

I believe better terms to use are:
sample rate -- measured in Hertz (or kiloHertz when that applies) for how often samples are taken of the input signal.

My English beeing a foreign language I have no really one-step name for the number of bits in a conversion as a concept, perhaps someone could help here. Sample depth is the word I tend to favor.

I agree that the terms are confusing.

I come from a family of engineers. My grandfather was a well-known recording/electrical engineer (and inventor) in Nashville working for Chet Atkins at RCA and Sam Phillps at Sunn. My father has a master's in electrical/industrial engineering and over 30 years experience in computer programming/engineering. They would be the first to tell you that engineers are a wacky group and always misuse/abuse terminology.

Most should know that referencing digital recording theory, Sample Rate and Bit Rate, are two very different things.

The best two terms to avoid confusion:

Sample rate is the common term defining samples per second (measured in hertz).

Bit Resolution is the common alternative (but still synonymous) to Bit rate in sampling theory. Both define the maximum bit word length used in Pulse Code Modulation (PCM).

As far as "selective dithering", that WAS the term used when it first came out years back. This was before there was much literature around, heard manufacturer Reps use it. As I mentioned previously in this post, I agree it is a confusing (poor) term. If there is a new term short of a technical description, do say. Unfortunately I really don't have the time or interest to seek out current literature on the process to see what folks are now calling it.

James,
we seem to agree. I still have problem with "bit rate", and you are probably 100% right in saying it is a common term. Nowadays, in the sound arena, I see it mostly in mp3 encoding, where you can select as example 128kbit / s as bit rate or perhaps a variable bit rate. Anyway, small point, it is friday and I´m off to other things.

Gunnar.

Thanks for replays. Now I'm searching for more tools.

Meantime, check this:

(dead link removed)

"The Old Yellow Board"

Topic: “Recording at 192 is a waste”

xaivious wrote: FYI - for those who are not familiar with selective dithering. It is suppose to select the empty (waste) bits to remove first. Those bits that contain silence and noise, maintaining the bulk of the signal best as possible. Normal dithering just chops 8 bits from the word (24->16) indiscriminately (sometimes taking the good and leaving the junk).

What you describe as "normal dithering" is incorrect. Indiscriminately chopping off the bottom bits (in audio, chopping the least significant 8 bits off of a 24 bit signal to obtain a 16 bit word) is called truncation. It is NOT dithering. And, it can lead to small DC offsets in the resultant signal and, more importantly, some correlated noise/distortion in the reduced bit length signal.

Take an 8 bit word: 11010000. After truncation to 4 bits it is 1101 - no error. Take 11011111. After truncation to 4 bits it is also 1101 - nearly a 1 bit error.

Dithering is the process by which a very small noise signal (below the least significant bit - in the above case the 4th bit) is added to the signal to smooth out the bit depth conversion. This signal can be white noise, but almost now is always spectrally-shaped noise (e.g POW-R).

Dithering will, occaisionally, push the bottom bit value up sufficiently to toggle the least significant bit of the bit reduced word and reduce the error that comes with simply chopping off bits. One way to look at it is a sophisticated rounding algorithm.

I have never heard the term 'Selective dithering'. However, a bit, in and of itself, doesn't contain silence or noise. The string of bits, formed as samples (16 bit, 24 bit, etc) over time defines what is silence or noise or signal - it's a lot more complex than that. LOTS

dpd wrote: What you describe as "normal dithering" is incorrect...

I have never heard the term 'Selective dithering'.

dpd,
This phrase was used to reference an opposite to a process that was at one point referred as "selective dithering", not the most common definition. If you read deeper into the discussion past the first couple posts you can see where the traditional definition of dithering is denoted, and what is being referenced there. In addition to some mention of the misuse/abuse/overuse of terms.

The truncation processes, as far as I know, are not all equal much like sampling processes. Are some companies doing something special with their process? It's at least what they claim. I have never done much research into it in detail.

I was just trying to generalize while sharing what I knew, not get into too many details. I personally am aware of the theory but there are likely plenty of folks who are reading this that are not. It is most definitely much more complex, and certainly not flawless either.