Hi, the title is self explanatory, but for more context I have 2 versions of an album which I would like to encode to lossy (AAC): The CD version with 44.1KHz-16bit, and the 2 channel SACD version with 88.2kHz-24 bit.
So, does a lossy encoding sound better/have better fidelity if encoded from the SACD source?

Or the extra "resolution" from the SACD source gets thrown away when encoding to lossy?

- OPUS  48kHz / 88.2kHz, depending on the program that analyzes it  :p

And for another question, someone has borrow me another album, this one in 5.1, also with 88.2kHz-24bit but obviously with 6 channels. How can I choose the bitrate?  I saw here a post talking about this, but I didn't understand it very well and honestly it was from 2006 or something, things may have changed radically meanwhile.

Psychoacoustic lossy formats have no bit depth per se - they store data differently and in floating point format internally.
Also they shall filter high frequency content above 16-22kHz, so, no - 88.2/24 source shouldn't give you any perceivable better fidelity.

Lossy encoders discard what you cannot hear - or more cleverly: what is expensive to encode and has small chances of contributing to fidelity. And that in particular goes for ultrasonics.
Moreover, SACD has extra ultrasonic noise, so encoding the ultrasonics would be extra costly for something that is even unwanted. Look at this https://hydrogenaud.io/index.php/topic,122580.msg1012188.html#msg1012188 . The noise rising from the 20 kHz frequency is not loud, but you do not want a lossy encoder to spend bits on it.
A lossy encoder that spends its bits on ultrasonics, would put fidelity at risk. And if you tried to encode that way - without the application overriding you - it would be worse with SACD. Of course if the encoder already has enough bits to spend on the music, it wouldn't be audible.
(...)
The extra "resolution" and the extra unwanted noise from SACD should both be thrown away. That is not to say that you will never hear the difference, if you fine-tune bitrate to the level where you with careful listening can tell one of them from the original - but then it is also likely to be due to differences in mix/mastering (i.e. the actual music).

This from 2023: https://hydrogenaud.io/index.php/topic,124479.msg1030864/topicseen.html
With a link to https://hydrogenaud.io/index.php/topic,120007.msg997612.html#msg997612 which sets forth a nice rule of thumb. In particular, if you want to encode 5.1, double the bitrate you would use for stereo.

Yes I know the ultrasonics are a waste for listening purposes (nor for mastering purposes I bet), but I thought it could be the case that lossy codecs had more to work with to improve the encoded sound, if sourced from those high sampled high bit depth files.

Musepack

while the latter is how 'close to analogue' you can make a digitally sampled wave.

while AAC & OGG when choosing 192 kbps ended up with bitrates generally associated with lossless files, of many hundreds of kbps

]$ oggenc -b256 "st.wav"
Done encoding file "st.ogg"
    Average bitrate: 253.2 kb/s

]$ oggenc -b256 "mch.wav"
Done encoding file "mch.ogg"
    Average bitrate: 268.5 kb/s

]$ ls -lh *.ogg
... 987K ... mch.ogg
... 932K ... st.ogg

you are going to keep the originals, right? Not throw them away ... ? You don't want to do a rip job more than once.
Then consider lossies as disposable.

What I do, is to encode to a format where I don't have much from other sources. Lossies are copied, lossless are converted: If my phone has an MP3 file, it is copied, not transcoded, but if my phone has a Musepack file, then I know it is converted from something and thus, this is not an "original" file.

Consider circles. You can reconstruct a circle just from 3 points. Having 100 points doesn't improve anything. Similarly for signals. You can reconstruct a band-limited signal from samples at sampling rate just higher than 2x the bandwidth of the signal. Having sampling rate significantly higher than 2x doesn't improve anything.

Quote from: pipoawas on 2023-09-04 19:01:52

while AAC & OGG when choosing 192 kbps ended up with bitrates generally associated with lossless files, of many hundreds of kbps

That's not my experience, at least with vorbis. Here "st.wav" is 2.0 file with pink noise and "mch.wav" is 5.0 file with pink noise, both 30 seconds long:

Code: [Select]

]$ oggenc -b256 "st.wav"
Done encoding file "st.ogg"
    Average bitrate: 253.2 kb/s

]$ oggenc -b256 "mch.wav"
Done encoding file "mch.ogg"
    Average bitrate: 268.5 kb/s

]$ ls -lh *.ogg
... 987K ... mch.ogg
... 932K ... st.ogg

 oggenc "filename.flac" -q 6

You're talking about the Nyquist frequency right?

But what I meant is, what if between each sample in 44.1kHz, those intervals with silence (no sampling) actually had some higher frequency, that granted we wouldn't hear, but if repeated throughout the music, in the long run we could actually "feel" in a non quantified way?

In the command line I used

 oggenc "filename.flac" -q 6

, which is the equivalent for 192kbps more or less. For some reason I couldn't make the --bitrate parameter to work.

Quote from: rutra80 on 2023-09-04 08:54:17

Psychoacoustic lossy formats have no bit depth per se - they store data differently and in floating point format internally.
Also they shall filter high frequency content above 16-22kHz, so, no - 88.2/24 source shouldn't give you any perceivable better fidelity.

Thanks, I didn't know about lossy not having bit depth.
As for the frequency, is that really it? I mean, lossy encoding shedding away the frequencies above 20kHz is different to the frequency of the sampling rate. The previous is how much you can hear a high pitch sound and discarding what you can't hear, while the latter is how 'close to analogue' you can make a digitally sampled wave. And since OGG seems to preserve that original high resolution sampling rate, does it mean the sound has more "warmth, fidelity, dynamics" and all that stuff that people swear they have with analogue (sometimes overselling it a bit)?
  Or do those high sample rate, high bit depth sound only matter at the mastering level?

Quote from: pipoawas on 2023-09-05 02:54:03

But what I meant is, what if between each sample in 44.1kHz, those intervals with silence (no sampling) actually had some higher frequency, that granted we wouldn't hear, but if repeated throughout the music, in the long run we could actually "feel" in a non quantified way?

Then writing about "how 'close to analogue' you can make a digitally sampled wave" was a really strange way of expressing it.

Well, there you go. "-q" is not about bitrate but about quality. It does automatically what the article that was linked describes how to do manually. It chooses a bitrate to keep some specified "quality". It shouldn't be a surprise that to keep the same quality with more channels, it needs to allocate more overall bitrate.

Many encoders apply bitrate dependent lowpass filter no matter the sampling rate.
Even lossless high bit depths and sample rates are practically impossible to double blind test against 44.1/16, so preserving such content in lossy encoding is considered useless - yes, their use is mostly for mastering or other special purposes.

Yes. If you can't ABX CD from SACD (assuming that mastering is the same), it's even less likely that you'll ABX lossy encodes of them. There's even a possibility of SACD encodes being inferior due to some unnecessary bits spent on ultrasonics in expense of what can actually be heard.

Filtering out something at 20kHz doesn't mean that precious bits won't be used for pretty useless 19kHz content.
The way data is stored in such lossy codecs makes it quite easy to resynthesize the sound at any bit depth and sample rate. Vorbis uses original sample rate to avoid resampling which may have its own share of problems.

But what I meant is, what if between each sample in 44.1kHz, those intervals with silence (no sampling) actually had some higher frequency, that granted we wouldn't hear, but if repeated throughout the music, in the long run we could actually "feel" in a non quantified way?

I know this is a somewhat controversial topic with lots of placebo and wishful thinking in it,

By the way I use terms like "feel" and warmth", because I'm kind of a noob at this level of audio knowledge

and not because I think vinyl is better "just because vinyl", since as long as the masters are good and the same between it and e.g a CD, in many instances the "warmth" of vinyl derives from it's amplifier, that "colours" the sound.

I see. More like a gimmick and "buy all this stuff again because theoretically is better".

If you can't hear it you can't "feel" it.   (Well....  You can fell deep or subsonic bass in your body but that's a different topic.)

And if something makes you "feel different" that can be demonstrated (or debunked) with a  Blind ABX Test.

It's only "controversial" among people who "don't believe in" blind listening tests.   (And unfortunately, the "audiophile community" is dominated by irrational people.)
(...)
A lot of "experienced audiophiles" use also these meaningless words.    Even common words like "warmth" have more then one meaning.   You could fill a dictionary with audiophile nonsense.  

Audiophoolery defines the 4 characteristics that define/determine sound quality.    (There are more than 4 "words" but the terminology should fall-into one of the 4 categories.)

A good amplifier won't have any audible characteristics unless you over-drive it into clipping (distortion).   Sometimes there is audible noise (hum, hiss, or whine, in the background) but distortion and frequency response (the only other two things that affect sound quality of an amplifier) are almost always better than human hearing.     Phono preamps (and microphone preamps) are high gain and are more prone to noise because every active circuit generates some noise and amplification amplifies the noise along with the signal.

Vinyl has several "limitations" but again, the noise is the most obvious and the most unavoidable.    The low-level background noise may be part of the "vinyl warmth".   

It's OK if some people prefer the sound of vinyl and that's a matter of taste.   But technically. it's inferior to digital.   And you can digitize vinyl and the digital copy will sound identical to the vinyl (in a proper blind ABX test).

I could build an amplifier that goes to 1 MHz.   (It's not that hard if you don't need high power.)   You could say it's "theoretically better" than an amp that only goes to 20kHz.    But for the purpose of audio it's NOT even "theoretically" better. 

Somebody once said, "The wider you open the window, the more dust comes-in".    And if you are converting SACD you should filter-out the ultrasonic noise because there is a possibility of it "causing problems" if you amplify it and send it to your speakers.  (I think most SACD/DSD DACs filter it out, but sometimes people convert to high-resolution PCM without filtering.)

What do you guys think about this video:  Sound beyond 20khz: The key to true high-fidelity?

For DSD64 to PCM, a target format of 24-bit 88.2kHz is enough. While there is some seemingly rise of noise above 20kHz, don't forget the noise floor of 16-bit is also 48dB higher than 24-bit, yet some CDDA releases still have noise shaping to pump up the noise floor beyond 16kHz or so.

The -1dB@25kHz.txt filter I made can be download here:
https://archimago.blogspot.com/2022/06/notes-on-dac-dsd-1-bit-pdm-measurements.html?showComment=1655906640637#c3453242716296098960

Decimation should be set to 8 regardless of output sample rate as "Auto" seems buggy