Audacity Forum

Steve:
OK, I'm convinced. Your example is excellent and clearly shows the effects of dither when processing tracks that are not at 32-bit floating-point. An interesting extension to this is to successively apply a Fade Out, then a Fade In, then a Fade out, etc. to the PCM tracks and check the noise floor after each operation (as interpreted from the "New Peak Amplitude" in the Amplify dialog when 50 dB of amplification is applied). Interestingly, the noise floor comes up a bit after the second operation (the first Fade Out), and stays constant after that - at least that's what I found when using Triangle dither.

Dep:
So the moral is, record on the Sony at 24-bit/44.1, set Audacity's Quality Preferences to 32-bit floating point, 44.1 kHz. Import your files - Audacity will losslessly convert them to 32-bit floating point. You can now edit, fade, amplify, compress, etc. to your heart's content and Audacity will do no "damage" to the files due to dither. Export to 16-bit PCM/44.1 - dither is applied once and once only, at this stage.

Despite what koz says about 44,100 versus 48,000 (which is true), you're going to end up at 44,100 for burning to CD anyway, so why go through the additional step of sample rate conversion, which may introduce its own artifacts.

-- Bill

billw58 wrote:Interestingly, the noise floor comes up a bit after the second operation (the first Fade Out), and stays constant after that - at least that's what I found when using Triangle dither.

OK, I take that back. The noise floor could increase with each application of an effect. My error was in using the fade-in/fade-out effect.

Instead, open the Nyquist Prompt and enter "s" (without the quotes, and nothing else). If I understand Nyquist correctly, this just returns the unmodified selection to the track. Applied to a 16-bit track, and writing down what the Amplify effect reports as the new peak amplitude after 50dB amplification (but Cancelling the Amplify), I get this (with Triangle dither) with each repetition of the do-nothing Nyquist prompt: -40.3, -34.3, -30.8, -28.3, -26.3, -24.7.

Does this make sense?

-- Bill

thanks very much, everybody.

here's what i intend to do: pretty much what i thought i'd do: first show at 16/44.1, bring it into audacity at that rate and depth, just for cutting and fading in and out and will then burn the result to a red book CD. the second show i'll record at greater depth, bring into audacity, and fiddle with it there.

i have an h2, which led me to get the sony -- the pres in the hr are kinda noisy, and as noted, the only analog gain control is the three-position switch -- the other control is after it was digitized, so it really doesn't do what i need a gain control to do. i use it now mostly for interviews and so on -- it's very good for that kind of work.

when i get the whole mess done, i'll report on the results.

<<<Despite what koz says about 44,100 versus 48,000 (which is true), you're going to end up at 44,100 for burning to CD anyway, so why go through the additional step of sample rate conversion, which may introduce its own artifacts.>>>

I agree, and it's a fine point between those two, but not so fine between 44100 and, say 96000. Whom do you trust, the anti-alias filters inside the capture device, or the software downconvert in post. Doing effects, even if it's only fades, can create damage when all the sound signals from 17 KHz and up are "magic."

-- From Sony...
-- The PCM-D50 is a 96kHz/24 bit linear recorder that records in standard .WAV file format and surpasses standard DAT sound quality --

Not surprising since DAT is 48000/16. It's the audio portion of a digital television tape machine. We have a DAT machine in the video machine room and it plugs straight into an edit conroller and will edit and switch with all of our tape machines and edit rooms.

I'ts almost never a bad idea to capture too much.
Koz

kozikowski wrote:Whom do you trust, the anti-alias filters inside the capture device, or the software downconvert in post.

With a high quality device like the Sony, I would expect that the ant-alias filters would be very good in either hardware or software. Even on cheap recording devices the anti-alias filters are pretty good these days thanks to mass produced A/D converter ICs.

The standard anti-alias filter in Audacity is provided by libresample and is very good, but not quite state-of-the-art, so the hardware in the Sony may be marginally better.
The alternative libsamplerate (which cannot be included as standard because of licensing conflicts with VST support) IS state-of-the-art. The differences between these two libraries is small but measurable, but I can't actually hear any difference.

kozikowski wrote:Doing effects, even if it's only fades, can create damage

As illustrated in the tests described earlier, but when there is only a small amount of processing being done the "damage" is minute - I would dare to say "insignificant".

kozikowski wrote:-- The PCM-D50 is a 96kHz/24 bit linear recorder that records in standard .WAV file format and surpasses standard DAT sound quality --

It's the analogue components that are usually the limiting factor in terms of sound quality. Even at "only" 44.1kHz / 24bit it's unlikely that the sound quality will be let down by the format. It would be interesting to do a double blind test of 44.1/24 vs. 96/24.

kozikowski wrote:I'ts almost never a bad idea to capture too much.

Except that going up from 44.1 to 96 kHz will reduce the available recording time to less than half, which may be an important factor with a portable recorder.

billw58 wrote:OK, I take that back. The noise floor could increase with each application of an effect. My error was in using the fade-in/fade-out effect.

Yes - I spotted that. Each time you apply a fade-out-fade-in you are reducing the peak audio level by 12dB (ignoring dither).

billw58 wrote:I think the "during recording" qualifier is needed to differentiate the three things that are happening here (as I understand it so far ). 1. All internal processing is done at 32-bit floating-point regardless of other settings. 2. Recording is done according to the Quality Preferences setting. 3. When importing, the highest available bit depth is used: either the bit depth of the file being imported, or the Quality Preferences setting.

Have I (finally!) got this right?

I believe that is correct. My tests certainly seem to agree.

billw58 wrote:I'd combine a modified version of what's in the manual with your expansion on it, as follows:

Audacity uses 32-bit floating-point samples internally while you are working on a project and exports your final mix using 16-bit integers.....

That all looks pretty good to me, except for a small detail that Export format is not necessarily 16 bit. In Audacity 1.3 you can select other Export formats that may be greater or less than 16 bit. In the context of the rest of the paragraph it appears to be taking about recording for CD, hence the reference to converting back to 16 bit, but I don't think it is very clear. I've updated the page (with a less wordy version) and notified the relevant people to ask for it to be checked for accuracy.

dep wrote:when i get the whole mess done, i'll report on the results.

I look forward to hearing how it goes.

it's my intention to do close to a comparison between 96/24 and straight 44.1/16 by using one on one night and the other on the other night -- it's a two show weekend. it won't be exactly a:b because the rooms will be different.

the whole thing arises from a studio album just done, which i think sounds horribly overprocessed. my thought is that the simplest path to CD will produce not just a very honest recording but one that is, too, the pleasantest to the listener. so much of my goal here is to avoid the "oh yeah, but in post you did this" phenomenon.

in the course of it, i've again discovered how much things have changed since my radio news reporter days in new york a zillion years ago. then it was an EV-635 feeding a sony cassette machine and more often than not feeding it via a voice-act over a phone line to the raddy-o station. (some of us tried tricky stuff, such as feeding at half-speed so as to double the effective bandwidth. which worked if you didn't mind losing everything below 200Hz.) the worst portable digital stuff is in many ways better than the best portable tape stuff of a generation ago. (anybody want to buy some really good marantz and sony portable cassette stuff? how about a 5-inch uher reel-to-reel? didn't think so.)

anyway, my hope is to do as little as possible to the first recording -- half-second fade in, 1 1/2-second fade out at each cut -- and have a high old time for myself with the oversampled, bit-deep second recording. oddly, it's the first -- adding the fades while changing nothing else -- that raises the question.

dep wrote:
anyway, my hope is to do as little as possible to the first recording -- half-second fade in, 1 1/2-second fade out at each cut -- and have a high old time for myself with the oversampled, bit-deep second recording. oddly, it's the first -- adding the fades while changing nothing else -- that raises the question.

Have you thought of doing cross-fades during the applause? Just my preference, but live recordings with fades sound wrong to me. You could create a seamless 80-minute set.

-- Bill

xfades are great, and i'd do that if the project allowed it. but the goal here is to make a CD that allows navigation by cut, and so it will need to be in pieces. if i can find a clean place to cut in at the beginning, i'd prefer that to fading in for the most part, though a little bit of fade will be needed to avoid the jarring appearance of the room noise. (this issue is particularly hellish with video, because just about *any* change of shots is going to alter the audio, creating an at least subconscious "what was that?")

in the final analysis, the content is going to dictate it, at least to some extent.

i have a CD of geoff muldaur, recorded in 1999 in germany, that very much achieves the effect i'm seeking. they did everything right -- except getting into and out of the cuts, which sounds as if they simply hit "play" and "pause."

certainly i want to get usable material, but this is all going to be an interesting experiment, too. with luck the result will be fit to give to a few of our more favored radio stations, so they'll have something in addition to the studio albums. but i'm getting <i>way</i> ahead of myself in even suggesting that. still, it would be nice to have something to send to friends who happened to be at the shows. to do that, i need to make the production invisible, by mostly eliminating the production!

Thanks everyone for the discussion about the Digital Audio page in the Manual (mostly Dominic's work). Steve made an edit to it along the lines discussed here, and asked me if it was OK. Well, it's OK as far as it goes, but then (I think) the original simpler version was. The original said 32-bit was used internally by default then 16-bit was used for export by default. That's correct. Default quality on a newly initialised Audacity is 32-bit, and default export is 16-bit PCM WAV (or AIFF on Mac, I think). Other inferences could be made than those made here about what happens if you import a 24-bit file at 24-bit quality.

The problem with trying to explain the whole subject in that paragraph is that it is even more complex than laid out there now because the behaviour is different for different importers. Here is what happens importing a WAV file via the libsndfile importer (this will be used if you import using the WAV filter or the "all supported files" filter):
I am not sure why we import a 24-bit file with 24-bit quality setting as 32-bit; you might argue for either of the numbers in brackets instead.

When we look at import of other formats:

* OGG is always imported at 16-bit irrespective of the file format and the quality setting.
* FFmpeg seems to import at 16-bit irrespective of file format and quality setting except that it will import bit formats below 16-bit as 32-bit float if the quality setting is 32-bit float .
* For QuickTime on Mac, the import respects the file format irrespective of the quality setting if that format is 16-bit or higher. If the format is less than 16-bit, it gets imported as 32-bit float.

Then you have the issue with what happens with recording. The recording may be expanded to the sample format in the quality setting, but people won't be recording in 24-bit with 16-bit hardware. Nor, as we currently understand from looking at our code, will they be recording 24-bit with 24-bit hardware into Audacity on Windows.

One minor quibble with what Steve said:

1) Create 3 tracks (it does not matter what the default quality settings are)
2) From the track drop down menu (click on the track name), set the first track to 16 bit, the second to 24 bit, and the third to 32 bit.
3) Select all 3 tracks and generate 30 seconds of silence..., [fade in and the 16- and 24- bit tracks will have dither noise]...

That's not true for the case where you had created tracks in 16-bit quality. In that case after both changing the track sample formats and fading in, the tracks are still silent because they can only be upsampled from 16-bit, so no dither is applied.

All in all, I'm inclined to revert the Manual back to almost what it was, apart from suggesting people read the Quality Preferences page for more detail. I think that's the place to (try and) explain all this.


Gale

<<<it was an EV-635 feeding a sony cassette machine>>>

Electro-Voice 635a, yes. I still have a 635a, I don't remember where I got it from, but the tape machine's long gone. I do have an original adapter cable, probably built in the early 70s. It was in use in one of two college ration stations. WMUC Maryland University or whatever the radio station is in Drexel.

I'll have some warm milk and lie down now.

Koz