Yes I know, thank you. Could you please tell me what bit "depth" do you read from the attached test file? It is the one where CoolEdit Pro shows 16bits actual bit depth and Bitter shows 32bits (I understand Bitter only shows bits pushed out, not the real resolution) and that was used in the test. It was not coded under Cool Edit, only analyzed.
-
Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Dithering is a Mathematical Process - NOT a psychoacoustic process.
- Thread starter j_j
- Start date
Yes I know, thank you. Could you please tell me what bit "depth" do you read from the attached test file? It is the one where CoolEdit Pro shows 16bits actual bit depth and Bitter shows 32bits (I understand Bitter only shows bits pushed out, not the real resolution) and that was used in the test. It was not coded under Cool Edit, only analyzed.
OP
- Thread Starter
- #363
Sorry, no. An appeal to ignorance remains a fallacy, and a dismissal of small variations in a reading does not impeach the entire idea. A substantial ISP of any amount is going to show up on any proper meter, and one that does at least 4x oversampling will be very, very close to the "true" measurement, so spare me this unwarranted, dangerous dismissal that can terribly, HORRIBLY destroy your audio.
So, then, any time you use a digital volume control that has steps other than 6.02dB steps (1 bit, exactly) you have irrevocably assured harm (which may be minimal in may cases, yes) to your original signal.
If you are doing this to avoid improper recording (i.e. any substantial ISP) then you are simply confounding one horribly improper act with another that may be pretty horrible itself.
ISP's of more than one percent or so should never, ever, under any circumstances whatsoever, be provided to an end user. Not now, not ever, never.
Refusing the obligation to produce properly is not an excuse, it is an offense to music, science, and mathematics.
I am hugely surprised an expert like you mention things like that. Isn't the thread starts with the benefits of dither? when dither is used, tiny volume adjustment is possible and the limitation of 6.02dB doesn't apply. The additional noise introduced by dither, say, 24-bit, which is supported by extremely cheap Realtek codecs, as well as high end 32-bit DACs, is negligible.
The fact is, many software player, DAC and other digital audio devices have build-in digital volume control, when people use them, even not for the reason of clipping, will change the source bit sequence anyway. So do you mean digital volume adjustment during playback/broadcasting should never be used?
xr100
Addicted to Fun and Learning
[Source: MP3 AND AAC EXPLAINED (Karlheinz Brandenburg.)]
IOW... forget asking what the "resolution" (or rather bit depth) is of MP3. ;-)
Or 16 bit LPCM for that matter, 16 bits is not the "resolution." (See examples in this thread of dithered 16-bit w/ 1kHz sine 10's of decibels below -100dBr.)
You can see my post comparing the spectral error (averaged over 10 seconds) of 320kbps AAC and truncation, although that could be misleading on its own without much further qualification.
And note:
Last edited:
xr100
Addicted to Fun and Learning
From the iZotope link:
"To show that true peaks can become arbitrarily high, we’ll explore a pathological waveform where we can make the true peak as high as we want, by adding more samples."
But then, such a "pathological" waveform would be managed by any "True Peak" limiter. The fact that one might come up with "True Peak" a fraction of a dB different to another is neither here nor there in this context. And, by allowing some margin by restricting to slightly below 0dBTP, all should be well...
But that is exactly the problem, what happens after "content delivery" is unknown. The user could have a DAC that has some headroom above 0dBFS, allowing for True Peak excursions above, or they might not. The source could be lossy compressed before final content delivery, or by the end user. It could be resampled before final content delivery. Etc...
Peak modulation in a delta-sigma DAC tends to be very considerably below 100%. Some DAC's operate are with excursions some dB above 0dBTP, as some margin is available for this.
(BTW, there have been cases where the digital volume control in a finished DAC product was not implemented correctly, and at the maximum setting clipped.)
Don't forget that clipping can occur in the analogue stages even if the modulator is happy.
More advanced users can make informed decisions. Personally, I allow for some margin from 0dBFS sample value before the signal hits the DAC, the gain is adjusted in 64-bit float and then dithered for delivery to the DAC.
But try asking the "man on the street" what 0dBFS is... let alone this "True Peak" business...
BTW, you mentioned Realtek? Last time I looked at their datasheets, IIRC the products to which the datasheets pertained used (on-chip) analogue volume controls.
Last edited:
For example many Youtube videos use the Opus codec and the codec always resample to 48k.
Well I meant Realtek products in PC mainboards. So in a PC environment, software volume controls are usually available in the playback software and OS mixer. For example, here is a video showing how to avoid clipping in a software player, with a Realtek codec:
https://hydrogenaud.io/index.php?topic=114125.msg940028#msg940028
In fact, when I mentioned ISP in the Reaper forum, some members thought that I was a shill who sells meters and plugins:
https://forum.cockos.com/showthread.php?p=1939405#post1939405
Last edited:
Full-scale signal to noise floor per IEEE 1057/1241/whatever the DAC one is (I always forget, 1272? -- I mostly did ADCs and was not involved with the DAC standard, just the transient recorder and ADC). You integrate the noise floor to get SNR. That is also for an ideal quantizer, quantization noise only, and sine wave input. The SNR derivation is not too bad but the SFDR derivation gets messy, Bessel functions and all that jazz, and I don't recall if there is a general closed-form solution. The derivations are not based on FFTs, which can limit the SFDR (assuming it is not limited by something else, like circuit noise and nonlinearity, so FFT size and sampling rate are not usually the problem in the real world).
I am certain you know all this, maybe my definitions differ from audio norms?
xr100
Addicted to Fun and Learning
In creating/uploading your own videos to YouTube, that's a known target. YouTube also uses replay gain, I think?
Imagine a scenario where a CD is released, a copy is acquired by @restorer-john, and played on one of his old multi-bit DAC CD players...
Realtek ALC898 HD Audio Codec.
To quote:
"Hardware Zero-Detect output volume control
0.75dB per step output volume and input volume control"
From the datasheet:
So the Realtek you tested has some margin for ISP's? Good.
To be fair, the problem has been tackled from multiple directions. All up-to-date commercial software limiters adhere to "True Peak" anyway. Even old versions of Waves L1 have an "Analog" mode, though the overhauled "25th Anniversary" version adds "True Peak" as well.
Also, remember, most of the stuff in the BS1770 standards is about consistent ("perceived") levels for broadcasts. The "True Peak" measurement is one aspect of overall loudness management.
I did the volume reduction in software before sending to the Realtek, so whether the Realtek has hardware headroom or not is unimportant.
Some videos are loudness adjusted, some not, the algorithm being used is a mystery, and a moving target. They can always change the rule, they can even allow more gain on premium members to make their contents louder, who knows.
In this case, older CDs don't use true peak meters during mastering anyway.
xr100
Addicted to Fun and Learning
We seem to be talking about different things. The volume can be managed in software, no problemo.
Right. Which could be a problem...
So, should a mastered CD have "True Peak" levels exceeding 0dBTP? The answer is NO, because CD has more than ample dynamic range, and what happens to the LPCM data on the CD subsequently is unknown.
https://www.waves.com/plugins/wlm-loudness-meter
Current price in USD is $49.* For content created on a commercial basis, there is no excuse...
(The "PLUS" version of that plug-in, which is included, also incorporates a "True Peak" limiter.)
And in the post-physical media world...
https://artists.spotify.com/faq/mas...-is-loudness-normalization-and-why-is-it-used
(*Applying a further 30% off using the discount code advertised on the product page.)
Last edited:
Use dBTP metering or not is up to you, if you are the mastering engineer. For consumers, they have what they have, so the only practical way to avoid clipping is to to use the digital volume control.
xr100
Addicted to Fun and Learning
Right. (Well, sort of, as you can always use "True Peak" limiting.) I'll explore your videos properly later, might be interesting or handy.
Of course, we haven't even considered other acts of stupidity in mastering (or earlier in production)--like sending out to a D/A-A/D loop for the purpose of slightly clipping the inputs of the A/D...
Last edited:
OP
- Thread Starter
- #374
Yes, you can dither THEN YOU ARE ADDING NOISE YOU DID NOT HAVE TO ADD, because you SHOULD NEVER HAVE USED THE DIGITAL VOLUME CONTROL ON A SIGNAL. You have now accused me of malpractice, and I require an apology and a retraction.
I am quite aware of the terrible gain profiles of many systems. Excusing that is ridiculous. Fix the gain profile instead. Don't use a digital control unless you really don't care about noise floor. Geeeze. Elementary engineering there, sport.
And don't add ISP's. That's simply inexcusable, and means that somebody, somewhere, will have to add noise to reduce the level. Just don't.
Finally, your fallacy of the excluded middle, it's not "never or always" add digital gain, quite obviously to any REASONABLE PERSON WHO DOES NOT FALSELY DEFAME OTHERS there's a middle ground. Don't do it when you don't have to (in production you'll have to, we're all clear on that, I think), and DO NOT ASSUME THE CONSUMER DEVICE DOES THE RIGHT THING.
That's a bad assumption. Don't add more quantization noise if you don't have to.
OP
- Thread Starter
- #375
OP
- Thread Starter
- #376
A lot of people can't understand what an ISP is. That does not excuse their existence. As to "PC Environment" there is a long story there, and the gain structure is profoundly insane.
OP
- Thread Starter
- #377
At the cost of adding more noise, of course.
Exactly. Things like perceptual encoding, EQ, etc (which will also add more noise, as well) may happen. EQ can cause many disasters in systems that pay no mind to gain structure, too.
But the problem remains the same. If you're producing a track, and you want it to come out the same everywhere, don't put in any ISP's.[/quote]
OP
- Thread Starter
- #378
Sorry. The mastering engineer has the ability to avoid this. The consumer should not have to compensate for the production.
xr100
Addicted to Fun and Learning
Alas, not possible in a lot of software.
But just in case it's of use to anyone, here's a plug-in that can do it (FREE):
Airwindows - BitShiftGain.
(Note that this is not an endorsement or lack thereof for the rest of the software offered by Airwindows. Suffice to say that they offer lots of plug-ins of varying utility.)
Last edited:
xr100
Addicted to Fun and Learning
True.
If, for instance, you're doing, say, DSP loudspeaker management (i.e. EQ, crossovers, etc.), then you've got to set the gain at digital output to allow for some headroom toward 0dBFS and, of course, dither back to fixed point for delivery to the DAC. I don't think one need be overly concerned about adding noise with 64-bit float processes, FIR (of whatever phase/delay characteristic) filters can be used where appropriate if really paranoid...
Of course, this is an example where you "have to" adjust gain digitally.
And, obviously, a great many consumer products are now doing lots of DSP before hitting the converters (soundbars etc.) Of course, one might suspect that it is often not done "right"...
Last edited:
Similar threads
