Another sample rate/bit depth question...

[JumpingJackFlash]

Is it best to compose, record and mix in the highest sample rate and bit depth available to you, then dither it down to 44.1/16 to put on a CD, - or, is it better to do everything in 44.1/16 from the get go, so no dithering is needed? - Which gives the best end result, or is there no difference?

Also, is there a relationship between bit depth and sample rate? - For example, I hear many people talking about either 44.1/16 OR 48/24. - But can you, for example, have 48k, and only 16 bit? - Are there any advantages/disadvantages to doing this?

Finally, I know an audio CD can only take 44.1/16, but if, for example, you render your music to a video to put on a DVD, can that take higher rates/depths?

Thanks!

[lindseyp]

disclaimer: I'm no recording engineer, but as an "Audiophile" and software (read microelectronics and electronics) engineer I have studied signal theory...

Sample rate:
if you wish to mess with the pitch afterwards...

AND

you have a *really* good bitrate converter (read- expensive!!!)

then fine, record in 96khz and convert to 44.1 later.


If you don't mess with pitch - no need.
If you don't have an expensive converter - your loss.


Bit Depth:

Only worthwile if you use extensive compression or dynamic range trickery. OK we all do , I think the word 'extensive' is the key here.

...my2c

[audiobot202]

I've had similar questions recently due to a sound card upgrade. Try this thread:
http://www.kvraudio.com/forum/viewtopic ... 791b705e33

There's some useful information there.

In general, from my understanding, recording in 24 bits gives your mix more headroom. That is, the extra bits allow for more amplitude information and so dynamic rangte is more accurately represented in the mix. Having tried it myself in Tracktion, it almost sounds as if you have a lot of volume to play with and more sounds can fit into the mix more comfortably together. It also defines quiter parts more thoroughly and preserves detail.
Higher sampling rates, generally relate to preserving higher frequencies, but the advantages have been debated hotly. In general it seems that recording in higher bitrates, as long as you dither when you go back down to 16 bits, is considered more important.

[lindseyp]

OK I have to get technical now...

higher bit depth (i.e. 24 bit instead of 16) will give you more difference between "quiet" and "loud".. which is great if you want to record something which .. when listened to goes from "quiet viola solo" to "bone crushing full-orchestra chordgasm." or "quiet mouse" to Horde of Demons" type cinematic thing.

however this is all lost when you convert to 16 bits, all the benefit you retain is the clarity of difference in volume from when you did any effects or mixing. I'd argue no point if you're going to convert to 16 bit anyway.


As for the bitrate...


Anything higher than the standard 44.1khz will not give you any better sound (that you can hear!), but if you intend to work with pitch or ANY kind of effect that involves pitch, you may be introducing aliasing which *is* audible.

If that is the case you should work in the higher bitrate, then convert down.

BUT

Most converters use 'fast' algorithms which are not true to signal theory and actually can degrade the sound. To do it properly you must convert to a continuous signal or high bitrate and down again to take advantage of the Nyquist-Shannon sampling theorem's assertion on retention of sample quality. In practice we approximate it, but the level of approximation differs greatly by algorithm. Good ones are not cheap.

[peteblues]

you keep mentioning 'bitrate' which is confusing. The proper term is 'samplerate'. I personally don't need more than 44.1 kHz or 48KHz, but there is a good reason why to use more; plugins, especially samplers often have to compromise a lot with their resampling engine, and use faster converters, which can introduce aliasing. If you process your compositions at 96kHz, you improve the headroom, and then you can later downsample with a high quality sample rate converter to your desired sample rate (i.e. 44.1kHz for CD)
About the bit depth: In the digital world there is a lot of processing going on in the filters and additional quantization noise can occur, if your working with lower bit depths. So, basically, the more bit depth you can use, without compromising your CPU performance, the better. 32bit floating point processing is very acceptable.

Peter

[IIRs]

There are a lot of very wrong answers in this thread..

In no particular order:

is it better to do everything in 44.1/16 from the get go, so no dithering is needed?

It is not possible (or desirable) to keep your mix at 16 bit integer from start to finish, as your DAW almost certainly mixes it using 32 bit floats. So you have to dither your final mix anyway.

Also, is there a relationship between bit depth and sample rate?

They each represent resolution, but on a different axis. High sample-rates give extended frequency response, while high bit-depths give extended dynamic range.

however this is all lost when you convert to 16 bits, all the benefit you retain is the clarity of difference in volume from when you did any effects or mixing. I'd argue no point if you're going to convert to 16 bit anyway.

Actually, if you record at 16 bits with headroom, and then restrict your dynamic range further with compression, you end up with a recording that has less than 16 bits of resolution. The only way to ensure that your final 16 bit version is really 16 bit is to work at a higher resolution at all stages until the final one.

If you process your compositions at 96kHz, you improve the headroom, and then you can later downsample with a high quality sample rate converter to your desired sample rate (i.e. 44.1kHz for CD)

No, sample-rate has nothing to do with headroom, it is bit-depth that increases headroom, and this is what makes a higher-than-than-target bit depth more important..

[peteblues]

I meant frequency headroom, not dynamic headroom. Since, at 96kHz, the average human can only hear up to 1/4 of that frequency, even if there is aliasing in between 24kHz and 48kHz, chances are you can't hear it.

I guess, the term headroom can be generally very confusing.
I can adjust the headroom in my car....
...with tilt steering

Peter

[birrbits]

OK I have to get technical now...

higher bit depth (i.e. 24 bit instead of 16) will give you more difference between "quiet" and "loud".. which is great if you want to record something which .. when listened to goes from "quiet viola solo" to "bone crushing full-orchestra chordgasm." or "quiet mouse" to Horde of Demons" type cinematic thing.

however this is all lost when you convert to 16 bits, all the benefit you retain is the clarity of difference in volume from when you did any effects or mixing. I'd argue no point if you're going to convert to 16 bit anyway.

I would say that there is a very compeling arguement to record in 24 bit even if your destination is 16 bit. Often people will use the headroom arguement (which is a good arguement). What I rarely see is the arguement for quantization noise and as an electrical engineer, this suprises me greatly.

Briefly, lets round some numbers numbers between 0 and 10, but lets only use integers, so if a number is 2.5 we call it 3 but if its 2.4 we call it 2. Lets say we had two (real) numbers 5.5 and 6.5, we round them to 6 and 7, then add them together and get 13. Now if we added the original numbers together this would only have been 12, so we are off by 1.
Oops I said numbers between 0 and 10, so we divide both results in half and get 6.5 and 6. Unfortunately we said the results had to be real numbers so we change the 'rounded' result to 7 which is greater than the original result (still) by a value of 1. This difference of 1 is quantization noise in our finished product. 1 on a scale of 0 to 10 (pretty significant).
But what about 24 bit and 16 bit quantization noise in a file mixed down to 16 bit?
Lets say we only used 8 tracks (8 mono waves) then we added them together and divided by 8 (yes this is somewhat of an oversimplification of the mixing process).
For 8 16 bit tracks mixed to 1 16 bit file the resulting quantization error can be a maximum of 4 bits. Restated, each sample can be off from original by as much as 4 bits!
For 8 24 bit tracks mixed to 1 16 bit file the resulting quantization error can be a maximum of 0.015625 bits. Restated, each sample is guaranteed to accurately reflect the original and the quantization noise is effectively 0.


still awake?

[IIRs]

still awake?

Just about!

Surely your arguement doesn't apply to the majority of us who are using apps with 32 bit internal resolution? Its losing resolution in the initial recording that is the problem surely..?

[birrbits]

Surely your arguement doesn't apply to the majority of us who are using apps with 32 bit internal resolution? Its losing resolution in the initial recording that is the problem surely..?

Yes that is correct, its the accumulated rounding error from the initial recording.
Another way to think of it is this, say you had an analog mixer & you mixed 8 instruments and then sampled the output at 16 bit, thats fine.
But if instead you sample each of those 8 instruments at 16 bits & then mix them digitally, whammo the accumulated rounding error can be off as much as 4 bits (every sample can be off by as much as 4 bits) from the sampled analog mixer version.
Also note that this is assuming a very generic mixing algorithm (add and divide), but it is worst case.

[C00kie]

I disagree with you, Birrman.

Suppose I have 8 mono recordings in 16bits. The recording levels were set carefully so the peaks are near 0dB so each track has full 16bits dynamic range.

So the level of all the tracks can be divided by 8 (crude mixing, but in practice you set channel levels to -12 or -18 also) and retain all details since it is not summed in 16bits but more likely in 32bits floats. Any host of this century will not do the mixing in 16bits, especially if VST's are involved.

So having the separate tracks recorded in 16bits will not mean that the mixed end result in 32bits misses any details. And the files are only half the size

[birrbits]

So having the separate tracks recorded in 16bits will not mean that the mixed end result in 32bits misses any details. And the files are only half the size

reread my previous post, we aren't talking about any errors due to mixing itself (32 bit or otherwise), but rather accumulated quantization noise (rounding errors). This is an arguement for recording (everything) at 24 bit, even if your goal is a 16 bit CD.

The recording levels were set carefully so the peaks are near 0dB so each track has full 16bits dynamic range.

The error I am showing is assuming that the full dynamic range of a 16 bit A/D convertor is being used.

[C00kie]

I re-read it carefully, still I don't think so.

Suppose the rounding error on a single track = 1 (the least significant bit) then divide by 8 (lower the volume) gives an error of .125 and that times 8 (number of tracks) is back to one in total again. Nothing lost, nothing gained. Now where did I go wrong?

[Shamanix]

I think the best idea is to export the final file with as high bitrate and samplerate as possible, then use a good dither (eg Ozone) and a good resamler (eg r8brain pro)

[birrbits]

I re-read it carefully, still I don't think so.

Suppose the rounding error on a single track = 1 (the least significant bit) then divide by 8 (lower the volume) gives an error of .125 and that times 8 (number of tracks) is back to one in total again. Nothing lost, nothing gained. Now where did I go wrong?

The error isn't neccessarily in the least significant bit.