Audio question I've never gotten a clear answer on...

[Debinani]

Okay, here's the question - maybe some of you guys with scads of experience with this stuff might know the answer:

Let's say I have a computer-based studio. Everything - and I mean EVERYTHING - I do is done entirely with softsynths - I never bring any audio data into the computer from the outside, period. Even if I had another workstation to render the synths I'd use FXTeleport or some other means to transfer the data over ethernet.

So, in this situation, do I get any benefit whatsoever from my sound card?

Is the sound card in any way used to process softsynths hosted as plugins on the same machine? Would a nice card let me run more softsynths or have more polyphony?

I'm trying to decide on a birthday present for myself and it's either going to be a new library or a new sound card - so I kinda want to know the answer. :)

[nlamartina]

If you're only rendering and not bouncing through the soundcard hardware, then no, it never travels through the hardware when going strait to disk. I could render a big ol' song in 24-bits with a computer that has a SoundBlaster 16 installed, and then take an identical project and render it with a computer that has a Presonus Firepod, and they will be absolutely identical to one another. Cases like what you're talking about use the card for monitoring purposes and nothing more. So while monitoring with a SB16 certainly may sound crappy, the program (let's say Vegas 5) is still going to render in 64-bit floating point, regardless of what the sound card is capable of. The sound card never comes into the equation.

However...

It's still a good idea to have a decent sound card because you really need an accurate representation of what your project sounds like so that your rendered file will be a faithful to your intended results. Monitoring in 16 bits is fine, but with 24 bits you get more headroom, more clarity, less aliasing in the final product, and (I think) most importantly, a more accurate stereo spectrum. Your end product is only going to be as good as what you can hear.

Finally, if you're using the MIDI interface on your card to drive your synths, a pro sound card is a MUST. You simply can't get away with some po-dunk DirectSound driver with a latency of 150ms. I don't care if it buffers it or not, because it not only causes your work process to slow down, but it also hampers the stability of your machine.

I hope I made your decision easier. :(

[Debinani]

That answered a lot of my questions, thanks!

I'm using an Audigy 2 Platinum ZS right now, so it's got ASIO drivers and all. Probably not worth it to bump up to a pro card then just for a slightly faster midi interface.

[Paulie]

I used to have an Audigy 2 Platinum ZS I ran it at 5ms latency without any problems at all. The midi interface is more than fast eneough.

The only problem with the Audigy is that you are limited in your choice for recording quality i.e. You can either use 16Bit at 48kHz or 24Bit at 96kHz and that's it. A lot of people like to work at 24Bit at 44.1kHz for which you would need another soundcard.

I eventually got myself an M-Audio Audiophile 192 which I have found to be superb for recording and mixing. I have also heard good things about the M-Audio 2496 which you can pick up dirt cheap these days.

[LowweeK]

You got everything in the replies, but I wanted to add another point : with dedicated and well optimized ASIO drivers, your CPU consumption will be lower than with unoptimized drivers.
I've read some tests done some years ago and at this time the Soundblasters where really CPU hungry compared to "pro" sound cards.

But now that we all have a high CPU power, I guess this detail is becoming insignificant.

[nlamartina]

You got everything in the replies, but I wanted to add another point : with dedicated and well optimized ASIO drivers, your CPU consumption will be lower than with unoptimized drivers.
I've read some tests done some years ago and at this time the Soundblasters where really CPU hungry compared to "pro" sound cards.

But now that we all have a high CPU power, I guess this detail is becoming insignificant.

This is absolutely true, and I'm glad you mentioned it. SoundBlasters are notoriously resource hungry, although the kX drivers were always a nice alternative to Creative's ASIO. Let's put it this way: With my old piddly machine, I couldn't get my Live! card below 20ms. 40ms was safe. I later bought a Presonus FirePod. 10ms was no problem on the old machine with it. I'm now at 1ms on my current workstation. Engineering dedicated toward music makes a big difference.

[PPH]

If you're only rendering and not bouncing through the soundcard hardware, then no, it never travels through the hardware when going strait to disk. I could render a big ol' song in 24-bits with a computer that has a SoundBlaster 16 installed, and then take an identical project and render it with a computer that has a Presonus Firepod, and they will be absolutely identical to one another. Cases like what you're talking about use the card for monitoring purposes and nothing more. So while monitoring with a SB16 certainly may sound crappy, the program (let's say Vegas 5) is still going to render in 64-bit floating point, regardless of what the sound card is capable of. The sound card never comes into the equation.

However...

It's still a good idea to have a decent sound card because you really need an accurate representation of what your project sounds like so that your rendered file will be a faithful to your intended results. Monitoring in 16 bits is fine, but with 24 bits you get more headroom, more clarity, less aliasing in the final product, and (I think) most importantly, a more accurate stereo spectrum. Your end product is only going to be as good as what you can hear.

Finally, if you're using the MIDI interface on your card to drive your synths, a pro sound card is a MUST. You simply can't get away with some po-dunk DirectSound driver with a latency of 150ms. I don't care if it buffers it or not, because it not only causes your work process to slow down, but it also hampers the stability of your machine.

I hope I made your decision easier. :(

Greater bit depth (i.e. 24 bits instead of 16) doesn't reduce aliasing. Greater sampling rate, however, does.

[nlamartina]

Greater bit depth (i.e. 24 bits instead of 16) doesn't reduce aliasing. Greater sampling rate, however, does.

Why do you dither then, going from 24 bits to 16 bits?

[PPH]

Why do you dither then, going from 24 bits to 16 bits?

Dithering consists of adding a little random noise to a sound before reducing the bit depth. When you convert a sound from, say, 24 bits to 16 bits, you are losing the amount of amplitudes you can represent. Each 24 bit number must be rounded off to the nearest 16 bit number.

This means many samples which many different consecutive numbers may be rounded to the same number. So, for example, it may happen that, in places where you had a waveform going slightly up and then down, rounding may cause a straight horizontal line to replace that curve. Adding a very small amount of noise to the signal before reducing bit depth can make the differences between similar numbers become more pronounced, so that when rounding-off occurs, these numbers are not rounded to the same number. This way, the irregular shape of the orignal waveform in that place is somewhat preserved. This is a very basic explanation, but I think it's not bad.

Now, aliasing doesn't have anything to do with that. Aliasing occurs when you record a sample that contains frequencies that are above twice the sampling rate. The digital to analog converter assumes all frequencies of the signal to be below twice the sampling rate. The numbers of a digital recording are like dots between which the DAC must draw a line. If a frequency is above twice the sampling rate, the DAC "thinks" they are other frequencies (aliases).

Sampling rate has to do with aliasing. Bit depth has to do with the accuracy with which you measure each number of the sampled signal. Each time you record a signal to a digital medium, a roundoff error occurs, because the original number must be rounded to a number that can be represented with the bit depth you have. The more bits you have, the smaller the error.

Now, theoretically, if you take the original signal and subtract the signal resulting from sampling it, you obtain a signal with the rounging off errors. Since the errors are somewhat random, the signal you would obtain from this operation is very similar to random noise, and it can be thought as noise. So, the rounding off errors produced when you reduce bit depth are perceived as noise. The smaller the bit depth, the more added noise is perceived in the signal.

This is strange, since dither is used to disminish the impact of reducing bit depth, and yet it adds noise... But that's the way it is :D I hope this wasn't too complicated. Please, someone correct me if I said something wrong or if I oversimplified it.

EDIT: I recommend reading the second chapter of the book which is freely available at www.dspguide.com. It's really well explained there. Chapter 2 has everything on the subject.

[nlamartina]

Oh duh! :( I should have thought about it, because it makes more sense. Thanks for setting me straight on that.

[White Noise]

Okay - here's one for you tech heads.

Let's say I'm a composer working on a broadcast brief that means I'm going to be working in 24-bit 48khz. I deliver my final mix to the dub fulfilling those criteria and it sounds as I composed it, however, from the same mix I burn a 16-bit 44khz CD copy to listen to on a stereo to check the mix. a) it's faster and b) it's higher in pitch. Question 1 is why? and question 2 is, how do I burn a CD copy of the 24-bit 48khx mix and retain the fairly crucial elements of pitch and tempo?

Thanks all.

[gsomers]

Okay - here's one for you tech heads.

Let's say I'm a composer working on a broadcast brief that means I'm going to be working in 24-bit 48khz. I deliver my final mix to the dub fulfilling those criteria and it sounds as I composed it, however, from the same mix I burn a 16-bit 44khz CD copy to listen to on a stereo to check the mix. a) it's faster and b) it's higher in pitch. Question 1 is why? and question 2 is, how do I burn a CD copy of the 24-bit 48khx mix and retain the fairly crucial elements of pitch and tempo?

Thanks all.

A lot of CD-burning software that finds as its source higher than CD quality sampling rate and bit depth files will perform the required down-conversion (both in sampling rate and bit depth) automatically (although the quality usually suffers somewhat compared to doing the conversion yourself before writing to disc). That's why I'm a bit surprised about your result. Evidently, your CD burning software is getting misled by something in the rendered 24-bit 48KHz file. And the way I'm thinking, it should actually sound slower and, perhaps, lower in pitch, because more samples (48KHz worth) are being misinterpreted into 44KHz space. In any event, try other burning software or, preferably:

Load up some audio waveform editing software (my examples will use Sony SoundForge 7.0 but every piece of DAW software worth its salt has equivalent functionality).

Start by resampling the 48KHz to 44KHz (on SoundForge 7 you go Process->Resample then select "New sample rate" of 44,100; "Interpolation accurary" of 4 for High; Select the "Apply an anti-alias filter during resample"; and make sure that "Set the sample rate only" is unchecked because checking it will get bizarre results not unlike those described in my first paragraph, above).

Next the bit depth of the sample must be converted as well (on SoundForge 7 you go Process->Bit-Depth Converter then select 16 bit in the "Bit depth" pulldown; Some "Dither" [try Gaussian]; and "Noise shaping" Off).

Save out the new file. That's it. The new file should be written properly to the CD as it is a 44KHz 16-bit encoded .wav file (you are using .wav, right - not .mp3 or some other lossy format?).

Please note that no matter what you do there will be loss of quality in two dimensions in your final CD result (you will have less dynamic range and less high end). I can usually detect a difference (sometimes it's downright easy to hear the difference). Others claim not to hear the difference (some people claim not to be able to tell the difference between a 128KBps bit-rate .mp3 file and a CD-quality .wav file).

Good luck!
Cheers,
George

[White Noise]

Thanks for all that George. Yeah, I use broadcase WAV since they allow me to tag the files with correct SMPTE timecode data as well.