Help - I've broken my Sinusoidal!! Dither it up please ?

[Madness]

Lastly, I would like to make something absolutely -clear- to everyone:
THIS IS A ""Tracktion-SPECIFIC"" PROBLEM.
I can run 28 of the same tracks in cubase sx 2, all at once and the cpu/disk bars barely go up at all - the disk bar peaks at one quarter usage!
Ditto in other DAWs.
Tracktion is the only one which has some problems with going over 10 24/96 tracks, and this is either an inherent flaw, or an incompatibility between my system and the way tracktion does disk-buffering. The problem is not resolved by updates or by changing shit on my machine-
if anyone from mackie sees this - I'd love for someone to fix it - because I ain't buying v2 until I can run 20+ 24-96k tracks without stuttering...
Goodbye all-
m@

Yeah, sounds like your right!! I didn't know you had no problems in SX.

That said, as always with PC's there are thousands of Variables that need to be taken into consideration, and no developer can account for everything (except Macs') and as a PC user it helps to get familiar with troubleshooting so you can atleast go half way to fixing any problems.

I would say the next step for you would be to Try installing the Tracktion 2 demo on another PC (as the problem is clearly some kind of hardware 'ish prob) of any spec, or a MAC and see how you go. As well as on an IDE in your own system so that you may be able to definatively say what the problem is NOT.

Then, if you really wish to still go with Tracktion and pay full price, you can log a support call with Makie and give them a real basis to fix your problems.

When I get back from Holidays in two weeks I will try running 10 x 96\24 tracks on my system and see how I go.

peace,

[SleepyP]

for Mr Purple, generally speaking > cache is better and < seek time is faster.

just wanted to throw out 2 comments on sampling rates:
-it's a little-discussed fact that the higer the sampling rate used, the higher the CPU usage and the lower the latency given the same buffer size. thus if you're one of those "get the latency really low!" people, you will have to run higher rates to achieve the lowest possible latency.

-this will vary considerably by plugin, but i noticed a radical difference in the sound quality of some VST effects and instruments when going from 44.1 to 96 KHz. i couldn't believe how much better the amp sims i use sounded (seriously). this is not to say that magically by using a higher sampling rate all audio processing sounds "mad betta, yaul" but it CAN make a difference, and to my ears is usually worth the added CPU load...

[metamorphosis]

I would say the next step for you would be to Try installing the Tracktion 2 demo on another PC (as the problem is clearly some kind of hardware 'ish prob) of any spec, or a MAC and see how you go. As well as on an IDE in your own system so that you may be able to definatively say what the problem is NOT.

Then, if you really wish to still go with Tracktion and pay full price, you can log a support call with Makie and give them a real basis to fix your problems.

When I get back from Holidays in two weeks I will try running 10 x 96\24 tracks on my system and see how I go.

peace,

Cheers dude-

[IIRs]

record once at 44k, then a second time at 96k-
using a ***decent*** dithering algorithm (not talking ozone, goldwave, soundforge or anything else here - I'm talking cool edit/audition-level non-realtime dithering quality, here) dither the result down to 44k.

Eh? You dither the sample-rate? I suppose you sample-rate convert the bit-depth too?

FYI, dropping the samplerate requires a high quality samplerate converter. Dither is irrelevent unless you plan to reduce the bit-depth..

[metamorphosis]

record once at 44k, then a second time at 96k-
using a ***decent*** dithering algorithm (not talking ozone, goldwave, soundforge or anything else here - I'm talking cool edit/audition-level non-realtime dithering quality, here) dither the result down to 44k.

Eh? You dither the sample-rate? I suppose you sample-rate convert the bit-depth too?

FYI, dropping the samplerate requires a high quality samplerate converter. Dither is irrelevent unless you plan to reduce the bit-depth..

No, that's innaccurate. Dithering is: introducing random/shaped 'error' in order to produce more pleasing results when resampling from a higher resolution (temporally or amplitudinally) to a lower resolution.
Dithering -can- occur in picture resizing (amplitudinally), or in video-resizing (amplitudinally), or in video frame-rate-changing (temporally). Whether or not dithering occurs in the process of RESAMPLING is completely and utterly dependant on the "Resampler" in question...

In terms of audio, and in terms of amplitudinal resampling (bit-depth reduction/expansion) or in terms of temporal resampling (changing 'sample-rates'),
*Both* -can- involve dithering, the *only* difference is that you're dithering in the temporal spectrum, the other is dithering in the amplitude spectrum.
WHETHER or not they use dithering is down to the resampler in question. SOME resamplers dither the temporal spectrum, while others don't.
Don't try telling me otherwise unless you also have a degree in comp.sci and/or have studied sample theory. Musicians tend to obsfucate things a bit because they tend to understand how things work in different applications, but not what they're actually doing on a technical level.
Sorry if that offends anyone-
m@

[IIRs]

Don't try telling me otherwise unless you also have a degree in comp.sci

No I don't But I thought I had a good layman's grasp of it, so please tell me how random noise can be used to interpolate between samples?

My understanding of dither is not that is uses noise to "produce more pleasing results", but that it actually preserves information that would otherwise have been lost below the LSB level.. but when you are downsampling as in your example above, all the lost information is above Nyquist and can't be preserved.

If you know better, please explain..

[metamorphosis]

Don't try telling me otherwise unless you also have a degree in comp.sci

No I don't But I thought I had a good layman's grasp of it, so please tell me how random noise can be used to interpolate between samples?

My understanding of dither is not that is uses noise to "produce more pleasing results", but that it actually preserves information that would otherwise have been lost below the LSB level.. but when you are downsampling as in your example above, all the lost information is above Nyquist and can't be preserved.

If you know better, please explain..

Okay, it is beyond the scope of this discussion for me to explain sampling theory en generale, but you can find out what you need to know through google or wikipedia (hopefully).
Regards music there is this idea that using higher temporal sampling rates do not affect perception because the human ear cannot hear frequencies above 22k -
this is innaccurate, because although the human ear can indeed not generally hear frequencies above 22k, it can usually perceive the difference in terms of the temporal accuracy or 'resolution' of the samples, or 'smoothness' or 'transperancy' (or whatever you want to call it).
I speak for myself when I say that bass sounds better recorded at 96k/48k than it does at 44k - this is because more information is being stored about where the aliased nodes on the incoming waveform are at a given point in time. In other words, temporal accuracy has been increased - creating (in my perception) - a 'smoother' result. This has nothing to do with the -particular frequencies- (per se) than are being recorded.

Dithering does not preserve any -accurate- information- above or below nyquist (for example in the LSB's of the amplitudinal domain) - it just makes the result sound/look/feel like it has For example a 96k sample resampled to 44k using wave-shaped dithering will have a better chance (generally, though it depends on the algorithms used) of preserving the 'psychoacoustic sound' of the original than one resampled without wave shaped dithering. It either uses random bits or algorithmically-guided (wave shaping) decisions to place additional data in the output, that in many cases makes for a more pleasing or ""accurate"" (note the doubledoublequotes) sound/picture. This is easier to visualise than it is to hear, so I suggest you look for examples of picture dithering through google.
Here is a couple to start you off:
http://en.wikipedia.org/wiki/Dithering
http://www.grafx-design.com/01gen.html

And for the record - both bit-depth reduction and 'sample-rate' reduction are examples of audible 'resampling'.
Neither -has- to involve dithering-
they can both opt for straight conversion-
and the only reason why we refer to temporal-reduction as 'resampling' more often than we refer to bit-depth reduction as 'resampling' is because of convention and conditioning. BOTH change the sample data- they simply change it in very different ways.
m@

[IIRs]

Dithering does not preserve any -accurate- information- it just makes the result sound/look/feel like it has

I'm almost certain that is not correct. Each individual sample is not more accurate, but the average of many of those samples is. (Up to 4 extra bits of "statistical" information, according to Bob Katz).

I have also confirmed this with my own ears: a reverb tail can continue to "dissapear into the dither noise" long after its undithered version has dissolved into total noise..

BOTH change the sample data- they simply change it in very different ways.
m@

Reducing the bit depth can be as simple as throwing away the lowest bits. Dither can be used to statistically preserve some of that information.. or not.

Reducing the sample-rate can be as simple as throwing away every other sample.. (though you will need filtering to avoid aliasing of course) but if you want to convert from 96K to 44.1K you will need to interpolate new data inbetween samples. Are you telling me that dither noise is used to interpolate between samples?

[metamorphosis]

Dithering does not preserve any -accurate- information- it just makes the result sound/look/feel like it has

I'm almost certain that is not correct. Each individual sample is not more accurate, but the average of many of those samples is. (Up to 4 extra bits of "statistical" information, according to Bob Katz).

You've misunderstood what I've written - please read up on dithering.

I have also confirmed this with my own ears: a reverb tail can continue to "dissapear into the dither noise" long after its undithered version has dissolved into total noise..

I don't know what you mean by 'total noise' - where's the additional noise coming from? How is it being generated? By the effect? At any rate, it would surprise me if you can hear dither noise, given that it tends to operate at a low audibility (if it's done well), you must be boosting it up somehow-

BOTH change the sample data- they simply change it in very different ways.
m@

Reducing the bit depth can be as simple as throwing away the lowest bits. Dither can be used to statistically preserve some of that information.. or not.

No, that's innaccurate - you're not preserving any information - you're ---attempting--- to make it -sound- like you have. Look at the dithering pictures on those links-

Reducing the sample-rate can be as simple as throwing away every other sample.. (though you will need filtering to avoid aliasing of course) but if you want to convert from 96K to 44.1K you will need to interpolate new data inbetween samples. Are you telling me that dither noise is used to interpolate between samples?

It can be, yes. Depending on the resampler in question it may or it may not. Reducing bit depth can also simply be a matter of interpolation, or it can use dithering as well as interpolation, or it can be (as you said) "as simple as throwing away every other sample" (although in the amplitude domain the fundamental equivalent of this is to reduce the overall depth then fit the signal into that range by simply applying division to each sample and rounding up or down depending on the outcome of the sample - or you can use comparisons (a kind of interpolation) with data prior to and immediately after the sample in question to form a more accurate opinion of what number to change the sample to)- it just depends what level of psychoacoustic accuracy you're looking for.
Look, I'm not discussing this any more - everything you've mentioned above is -somewhat- accurate and what I've said is also accurate. Read those two links I posted and inform yourself on what dithering is- Cheers-
M@

[IIRs]

You've misunderstood what I've written - please read up on dithering.

I'm asking you to explain yourself better then. (I have read up on dithering btw: that is why I think you are wrong!)

I don't know what you mean by 'total noise' - where's the additional noise coming from?

Quantisation distortion from the truncation.

At any rate, it would surprise me if you can hear dither noise, given that it tends to operate at a low audibility

If I turn up my monitors I can hear 16-bit dither fairly clearly. In any case, it is easy to devise a test that can be listened to at more sensible monitor levels.

you're not preserving any information - you're ---attempting--- to make it -sound- like you have. Look at the dithering pictures on those links-

I did look at them. They confirmed what my ears tell me when i do my reverb tail test: there definitely is extra information in the dithered version that is lost in the undithered one.

Reducing the sample-rate can be as simple as throwing away every other sample.. (though you will need filtering to avoid aliasing of course) but if you want to convert from 96K to 44.1K you will need to interpolate new data inbetween samples. Are you telling me that dither noise is used to interpolate between samples?

It can be, yes. Depending on the resampler in question it may or it may not.

I don't have the knowledge to argue that, though it seems wrong..

Reducing bit depth can also simply be a matter of interpolation, or it can use dithering as well as interpolation, or it can be (as you said) "as simple as throwing away every other sample"

.. but that has to be wrong! What samples get thrown away when reducing bit-depth??

Look, I'm not discussing this any more -

[Samb]

Well, I can't really be arsed with another one of these sampling theory discussions and I don't have a degree in comp.sci (but I do have one in maths), but I'm just going to post here to say that there is no reason why a dithering process can't be used when resampling. Think of a graph with a waveform drawn on it, each sample on this graph is represented by two numbers a bit depth number shows its position on the Y axis and a sample position number shows its position on the x axis - there is nothing qualitively different in these numbers. Which resampling algorithms (if any) use dithering for resampling, I have no idea.

[IIRs]

Well. As far as audio engineers are concerned, whether or not dither noise is actually involved in the SRC process is pretty irrelevant, as it will be an invisible part of a function called "samplerate conversion". The only time an engineer or musician need be aware of dither is when reducing bit depth..

[Samb]

Actually, ignore my last post 'cos its complete bollocks. When I started writing it I was going to agree with you IIRs (mostly because I just didn't like metamorphosis's attitude that this was something that mere mortals were too thick to understand - I hate the attitude that maths is something arcane (and boring) and I think a lot of people are missing out on something beautiful, interesting and exciting:-o because mathematicians and maths teachers don't make enough effort to present mathematical ideas in an interesting way), but I talked myself round while I was typing and then posted without actually thinking through what I was saying

I'm going to think about it a bit more to make sure I get my facts right next time, but I think that actually you are right IIRs, that dither isn't (and can't) be applied in sampleraet conversion.

[valley]

Dithering is applied to bit depth reduction as an effort to reomve unwanted upper harmonics caused by the down sampling process[1]. The point behind dither is to try and blur those harmonics into the realms of (high frequency) noise. Hopefully the result gives a strong impression of the desired harmonics, and a (very) weak impression of the unwanted harmonics. The trick is that in this case the desired harmonics cannot be represented normally, so dither gives the impression of extra content.

When reducing sample rate however, the whole object is to ensure that *no* extra content makes it through the process, as this will always be interpreted by the ear a prominent, and usually unmusical, harmonic. It seems to me thefore, that the very purpose of a dither; namely to create an illusion of retained data, would, when applied to SRC simply introduce a differing form of alias distortion.

The point is that whilst mathematically, time and amplitude may be able to be treated equally, the effect that changes to either axis has on the ear is profoundly different.

[1] A triangle waveform can be reproduced using two bits of data, but if you bit reduce it to one bit, you have a square wave, which by definition has a whole lot more spectral content.

[Lunch Money]

Right, but a flawed argument. I'm not saying that there's not a difference-- I don't have enough expertise. I'm just pointing out an inconsistency in logic.

Let's say that it takes 8 bits to recreate a perfect sine wave. I mean, a perfect sine wave. Adding any more bits to that equation won't make it "more" perfect if it's already perfect.

Now, the reason I'm not arguing against the difference is that with music, we are NEVER talking about just a set of perfect triangle, square, or sine waves, so it's a moot point.

Greg