Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter? (Read 6210 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #25 –
Quote from: Porcus on
Honest question:
One of those plots show that the 88.2 kHz signal generates ultra-ultrasonic noise at 30 dB lower.
If that sort of noise is ever going to be a problem, presuming you don't deliberately play bat sounds at tweeter-splitting levels?
It may depend on amp and speaker design, and analog playback level, so can't give a definite answer without testing a lot of analog hi-fi equipment. Though I guess at worst it may just introduce more noise below 20kHz due to IMD instead of causing smoke.

The purpose of my previous post is to encourage a more rational thinking about ultrasonic noise level of software DSD->PCM converters, the fear of noise is completely hyperbole when compared to direct DSD playback.

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #26 –
30 seconds snippet for evaluation.

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #27 –
Quote from: akse0435 on
https://s-audio.systems/dsd-filter/?lang=en
Quote from: Larryg49 on
    http://s-audio.systems/dsd-filter/
Think I should test this one too.
X
The filter length is 729, longer than the other two filters (641), so will take more processing power. However, despite the longer filter length, noise rejection below 20kHz and above 50kHz is still slightly worse than the other two shorter filters. The S-Audio filter probably spend too much on the not very meaningful "passband ripple of 0.00001 dB".

Anyway, the differences are negligible when compared to direct DSD playback.

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #28 –
Quote from: forart.eu on
Well, I believe that the target-question should be: "How to correctly feed a DSD into a lossy encoder in order to obtain the best possible quality result ?"
That's another good way of putting it, yes  ;)

Quote from: DVDdoug on
  Just because you can hear pure-loud 20kHz signals in a hearing test doesn't mean you'll hear them in music.   And not everybody can hear up to 20kHz. 
(...)
But if you have too much frequency range sometimes you can get unwanted side-effects with subsonic or ultrasonic signals or capability...   You don't want subsonic or ultrasonic "junk" going through your amplifier and to your speakers

Two pertinent facts, thank you. Also, DC in audio? A continuous voltage? That isn't even audio anymore, why did the MP3 format allow that?

Quote from: bennetng on
The filter length is 729, longer than the other two filters (641), so will take more processing power. However, despite the longer filter length, noise rejection below 20kHz and above 50kHz is still slightly worse than the other two shorter filters. The S-Audio filter probably spend too much on the not very meaningful "passband ripple of 0.00001 dB".

What's that spike at 1KHz?  It can't be that most of the music is situated at around 1KHz range.
EDIT:  Got it, it wasn't music, just a sinewave test, at 1KHz   O:)


Thanks for your inputs guys.  :))

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #29 –
Quote from: pipoawas on
What's that spike at 1KHz?  It can't be that most of the music is situated at around 1KHz range.
EDIT:  Got it, it wasn't music, just a sinewave test, at 1KHz   O:)
I already posted some 30 seconds classical music audio files and plots in Reply #26 above, so you can check them out if desired. Perhaps you can do some ABX tests too.

When using music as input, the music spectrum will cover up most of the < 20kHz range so that it is impossible to check the noise floor clearly.

Also, the presence of a test tone can be used to align different plots. Without a tone, the spectrum can be shifted up or down in a confusing way which could be unfair.

In RMAA, every test has a different plot, so I only posted some of them to avoid spamming the page too much.

You can see an example of a more detailed report in the attached 7z file below:
https://hydrogenaud.io/index.php/topic,100481.0.html

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #30 –
Quote from: bennetng on
I already posted some 30 seconds classical music audio files and plots in Reply #26 above, so you can check them out if desired. Perhaps you can do some ABX tests too.
I tried to ABX them in foobar, but they all sounded each other transparent to me while I was pressing A and B, so I didn't bother voting  in which sounded better, because it would average at 50% of wild guesses .

I noticed that while the lossless -1dB@25kHz puts low volume ultrasonics up until 45KHz, the lossless Multistage puts a lot more dB ultrasonics in that area. Also interesting is that the lossy Vorbis Multistage out of the blue adds some strange ultrasonics from 40Khz up, in the last 5 seconds, that clearly shouldn't be there.
What software did you used, and is it a plugin for Foobar? I use Spek, but it's colors are less noticeable than yours, as you can see by my Multistage FLAC spectrum, specially the ultrasonic noise: 

(How do I post images inline like you did? :P )

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #31 –
Quote from: pipoawas on
What software did you used, and is it a plugin for Foobar? I use Spek, but it's colors are less noticeable than yours, as you can see by my Multistage FLAC spectrum, specially the ultrasonic noise: 
https://hydrogenaud.io/index.php/topic,123621.0.html
Quote
(How do I post images inline like you did? :P )
https://hydrogenaud.io/index.php/topic,111074.msg994047.html#msg994047
Whenever possible I often try to reduce the color depth to 8-bit when attaching PNGs. The problem is the very colorful SoX spectrograms are already 8-bit by default, but the graphical complexity still resulted in a rather big file size so I just posted them as thumbnails.

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #32 –
Quote from: bennetng on
Quote from: pipoawas on
What software did you used, and is it a plugin for Foobar? I use Spek, but it's colors are less noticeable than yours, as you can see by my Multistage FLAC spectrum, specially the ultrasonic noise: 
https://hydrogenaud.io/index.php/topic,123621.0.html
Quote
(How do I post images inline like you did? :P )
https://hydrogenaud.io/index.php/topic,111074.msg994047.html#msg994047
Whenever possible I often try to reduce the color depth to 8-bit when attaching PNGs. The problem is the very colorful SoX spectrograms are already 8-bit by default, but the graphical complexity still resulted in a rather big file size so I just posted them as thumbnails.

Double thanks  ;)

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #33 –
Quote from: pipoawas on
Also, DC in audio? A continuous voltage? That isn't even audio anymore, why did the MP3 format allow that?
MP3 frames are pretty short. They're short enough that you can encode a low-frequency (but still audible) signal and some frames will be indistinguishable from DC.

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #34 –
Opus hates subsonic frequencies. What I set in the screenshot below produced the attached audio files.
X

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #35 –
Quote from: bennetng on
Opus hates subsonic frequencies. What I set in the screenshot below produced the attached audio files.
[attach type=image]27241[/attach]
They all seem to up the frequency a bit compared to the subsonic original, not so much opus though.
But there was a lot of sound (noise) in the audible spectrum, and yet lame and vorbis failed to encode those, which they should, noise or not. Also opus put a lot more 'energy' near the ultrasonics and up.  What for?

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #36 –
While you can see more clearly with the SoX plots, the color schemes are not representative to hearing. Also, remember frequency perception is in log scale, not linear like what these graphs show. For example, if A3=440Hz, then A4=880Hz, A5=1760Hz and so on.

Also, even within the same format (e.g. AAC), there are different encoders with different implementations, even different versions of LAME encode differently.

The bottom line is don't try to force the encoder to do too many jobs on subsonic and ultrasonic, doing so may either introduce more potentially audible artifacts or inflate the bitrate.

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #37 –
henever possible I often try to reduce the color depth to 8-bit when attaching PNGs. The problem is the very colorful SoX spectrograms are already 8-bit by default, but the graphical complexity still resulted in a rather big file size so I just posted them as thumbnails.
Quote from: bennetng on
While you can see more clearly with the SoX plots, the color schemes are not representative to hearing. Also, remember frequency perception is in log scale, not linear like what these graphs show. For example, if A3=440Hz, then A4=880Hz, A5=1760Hz and so on.
What would be a better graphical representation of hearing in Sox?  I assume it's about playing with the -z dB parameter

Re: SACD to lossy vs. CD to lossy: do sample rate and bit depth matter?

Reply #38 –
I don't think SoX's frequency plot can be set to log scale or allow different FFT sizes, but in general, if an SoX graph's top dB range is 0, then 16-bit noise floor is at about -120dB. Now if you ask why it is not something like -96dB, then have a look on this article:
https://www.superbestaudiofriends.org/index.php?threads/tutorial-fft-size.5675/

In fact, if the SoX graph only has 8-bit color, then the bigger the dB range, the more coarse the graphical representation is. Let's say, if you set the range to 128dB, then each color can only represent 0.5dB. Human with normal hearing are more sensitive than this.

It is also pretty easy to realize that no matter how short or long an audio file is, the graphs are always at the same size, so the longer the file, the more inaccurate the graphical representation is. In the same way, the higher the sample rate, the fewer the pixels to represent the audible frequencies.

I think the latest version of Spek also contains some advanced options too, so you should try to figure out how to change the settings.
https://www.spek.cc/
Quote
Adjustable colour palette, DFT window size and function. The ability to switch between audio streams and channels.

For SoX, the manual also mentioned some other graph settings, so find and read it.

If you are interested in studying how human hear things and learn about how to use more different methods to analyze audio data, find some books or university websites that talk about psychoacoustics instead of simply asking on a forum.