Settings

Joseph Sinclair · **Voice Seeker** Joined: 20 Jan 2008 Posts: 17

I'm trying to set up a standard way of recording my real estate podcast (I'm not a voiceover person):

1. Record at 96 Hz/32-bit float (line-in from mixer)

2. Edit

3. Convert bit rate to 16-bit

4. Downsample to 44.1 Hz

5. Encode to MP3 at 96 kbps

I have a Sound Blaster Audigy 2 ZS Notebook (24 bit) card for my laptop, and I'm using Sound Forge 9.0.

Is this a good scheme to get a professional sound?

Comments? Advise?

Jacob Ekstroem · Posted: Wed Feb 13, 2008, 22:06 (GMT) Post subject:

Well.... why not just record at 16 bit/44,1 kHz to begin with? All that downsampling/bit-conversion seems a bit unneccesary, if your target output is a 96 kb/s mono mp3-file.... imho.

Joseph Sinclair · **Voice Seeker** Joined: 20 Jan 2008 Posts: 17

A few (not all) of the books I've read have suggested recording at 24-bit so as get more headroom to avoid clipping but still achieve a strong signal with low noise. I'm not sure how important that is balanced against the distortion resulting from conversion and downsampling.

This is all new to me.

Jacob Ekstroem · Posted: Wed Feb 13, 2008, 22:52 (GMT) Post subject:

Joseph,
what you have read is true, however... if you're completely new to the world of voiceover recording, I would say: don't worry about it. There are so many different aspects in making the perfect recording, that numerous books have been made about it. You have read a few of them, so you know. Learning and mastering all these aspects is another thing though, and it takes time and practice. I'm sure a 16 bit, 44,1 kHz recording (that's cd-standard) would work for you. If your signal chain is clean enough, you should be fine. I've done voiceovers for more than 10 years, and I've never felt the need to record in a higher quality than 16/44.

Of course, the best thing would be to do some experimenting. If you feel you master all the conversion and downsampling, and have no problems with the extra work, by all means record at the best possible quality. But, if your target is 96 kb/s mp3, I'm not sure even I would be able to hear the difference...

Anyway, actually I'm quite happy you have taken the time to investigate in this! That tells me you are serious - that's nice, and admirable.

Lance Blair · Posted: Wed Feb 13, 2008, 23:12 (GMT) Post subject:

Hi Joseph,

I hear a difference when I record 48k/24bit vs. 44/16...but when I send it to a mp3 under 128kbps...I might as well be recording 44/16, which should be absolutely fine for what you're doing.

Spend more time making sure the mic and mic use and placement is correct for you, and spend time making sure the room sounds good. That's more important than your recording specs for this.

Good luck.

Jacob Ekstroem · Posted: Wed Feb 13, 2008, 23:18 (GMT) Post subject:

Emmett Andrews · Posted: Thu Feb 14, 2008, 02:29 (GMT) Post subject:

No one on earth can hear the difference between 44.1 and anything higher. It's an impossibility, which has been proven time and time again. HOWEVER, some soundcard converters simply perform better at higher sample rates, which makes an audible difference. The thing to do is use your ears to determine how well your card performs at different rates.

As for bit-depth, I always use 24-bit (32 float). It's essential if you do any processing after recording. It's not that 24-bit sounds better, per se. No one can hear the difference as long as levels are set reasonably well. But it makes a HUGE difference in processing quality. Even just a volume change makes a 16-bit file sound a little nasty, and repeated processing can make is really ugly, really fast.

So, if you're not doing any processing, 16/44.1. If you ARE processing, use 24/44.1.

If I may ask though, why 96kbps? Why not 128 or even 160 (mono max)?

Emmett

Bill Campbell · Posted: Thu Feb 14, 2008, 02:55 (GMT) Post subject:

Emmett...nice info on the 32 float being better for processing. Is that true for Adobe Audition's 32 float setting?

Erik Sheppard · Posted: Thu Feb 14, 2008, 03:39 (GMT) Post subject:

Yeah, I was asking a friend about the 32 Float thing just the other day and he seemed confused about it as well.

Emmett, you make it sound as if it is the same as 24, is it?

Colin Campbell · Posted: Thu Feb 14, 2008, 04:15 (GMT) Post subject:

Emmett Andrews · Posted: Thu Feb 14, 2008, 04:38 (GMT) Post subject:

I'll explain it the best I can without being too technical.

Hardware cannot play back 32-bit floating point (or any floating point). It's the way numbers are handled internally. So, as many of you know, the bit-depth is how many binary digits can be used per sample. The higher the bit-depth, the more accurate the math. A 16-bit file holds 16 digits, a 24-bit file holds 24 digits. A 16-bit file offers 96dB of dynamic range, while a 24-bit file offers 144dB.

A 32-bit floating point file is a little different. It is a 24-bit file with an exponent. So, at the time of recording, it offers 24-bit quality. Processing, however, is MANY, MANY times more accurate. If you can remember basic exponential math, it looks something like this: 24^7 (the other bit is used elsewhere). This offers about 110,000,000,000 binary digits for each sample, and a dynamic range of over 1,500dB (I think I did my math right).

16 and 24 bit files have a digital noise floor that processes. Meaning, if you take a 16-bit file and normalize it to 0dB, you will have a blast of digital noise. The noise floor of 32-bit floating point does not exist, in any real sense. It is fixed at -150dB for playback, but cannot be normalized. You can amplify the silence all day and it will never raise above -150dB.

Using floating point technology, it is possible to exceed 0dB without clipping. It will clip the hardware, but it will not internally clip. Which means that if you add EQ that causes the file to clip, you can simply turn it down after processing. With a 16-bit file, once it's been clipped, you're left with a flat-topped wave.

And because of the extreme precision of floating point math, you can apply processing without any negative digital side effects. When using 16-bit, every effect comes with a digital flaw, known as a rounding error. The math has to round calculations, so it isn't accurate. So it is a trade-off between the effect you want, and the quality you lose. It's a grainy, nasty sound. Mild, but audible almost immediately. Floating point math has so many available decimals, that there is no need to round. Or if it does, it will be so incredibly minute that no one could ever possibly notice it.

Early on, someone explained it to me in terms of money. Think of sales tax rates...Where I live, it's something like 7.586%. Obviously, rounding must occur when I pay for something. At the end of the year, imagine how many extra fractions of a cent I've paid in sales tax. Probably enough to be a nice wad of bonus cash! Now imagine if our currency allowed for those small fractions of a cent. Over time, they would add up! And I would come out ahead! That's a rough analogy, but you get the idea.

And Bill, yes it's true of all floating point systems. Most current software uses floating point math. Pro Tools is the only modern system still using integer-based math for processing (that I'm aware of).

Emmett

Emmett Andrews · Posted: Thu Feb 14, 2008, 04:45 (GMT) Post subject:

Erik Sheppard · Posted: Thu Feb 14, 2008, 04:51 (GMT) Post subject:

Emmett, your floating point explanation was just incredible.

Thank you sir.

Chris Clementson · **Voice Seeker** Joined: 14 Jan 2008 Posts: 216

One reason digital sounds harsher than analog is clock jitter.

Todd Ellis · Posted: Fri Feb 15, 2008, 04:45 (GMT) Post subject: