[pinnahertz]

[/i]I know that I can't really hear stuff that are so quiet. But it does change what I hear.

Sorry, both cannot be true, it's one or the other.

I mean, the fact that the driver of my earphone is trying to make that sound does affect the overall sound, if you understand what I mean.

I know what you're trying to say, but I don't think you know what you mean. Or at least, you have an incorrect concept.

If you record a pluck of a guitar, and then cut out all sounds under -70 it will sound very different to the unedited version.

That depends on what you mean by "cut out all sounds under -70". I'm not even sure what you mean, or how that would be done, or why.

So I just want to keep as much info as possible.

Right. But you're going to m4a, so that actually dumps as much information as possible. That's the entire goal of lossless codecs, to remove as much information as possible and still have an acceptable result.

I
The color interpretation is different from Spek (the PC app I use), so take that into account when looking at dB. In Spek, for examle, bright green is ≈-65, here it's -50. The song it's quite. But the app is mobile, so it may be quite a bit wrong interpretating volume. My PC monitor broke, can't use Spek(

Really, do yourself a big favor and stop looking at spectrograms. They do not equate to audibility. If you study them and see a "huge difference" by your definition, you will strongly bias your perception of the actual result. As others have noted, they will display information clearly all the way down to the noise floor. Just don't do it.

 

[bigshot]

The spectral display looks very pretty.

 

[gregorio]

[1] Then I converted the 24/192 to 16/44.1, and for some reason, it turned out "worse", upper frequencies cut, less information, and it weighs less than the original 16/44.1 that I downloaded.
[1a] I know that more high frequencies doesn't mean better quality,
[1b] I just don't want to lose the airyness of some songs,
[1c] so I want to keep as much information as possible.

1. Why is it "worse" to have "upper frequencies cut, less information, and it weights less"? I don't know what you mean by "weights less" but "upper freqs cut" and "less information" isn't worse, it's better!!! Because of the following points:
1a. Correct. In fact, more high freqs can often mean worse quality as it's liable to cause distortion (IMD).
1b. "Airyness" exists in roughly the 8kHz - 16kHz range and mainly between about 10kHz-14kHz. Therefore:
1c. Why would you want to keep information significantly above that range?

[1] I mean, the fact that the driver of my earphone is trying to make that sound does affect the overall sound,
[1a] if you understand what I mean.
[2] If you record a pluck of a guitar, and then cut out all sounds under -70 it will sound very different to the unedited version.
[2a] So I just want to keep as much info as possible.
[3] I can't just download m4a files, because they are usually mastered for iTunes and sound funky.

1. But that's NOT a fact, the driver of your earphone is NOT trying to make "that sound"! For example, let's say you have sounds down at -80dB, the noise floor of the recording is say -60dB and you're listening at a peak level of 70dBSPL, your earphones are not trying to make those sounds down at -80dB, they're just making noise that does NOT affect the overall sound.
1a. Not really. Exactly what the driver of your earphone is "trying to make" is dependant on a number of factors. Additionally, A. "Trying to make" and actually making are not necessarily the same thing (IMD for example) and B. Even if they succeed and the overall sound is affected, that's no guarantee that the difference is in any way audible.

2. That is simply NOT true. If one were to cut all sound below -70dB in the vast majority of reasonable reproduction scenarios it would sound absolutely no different whatsoever, let alone "very" different! Even in some fairly extreme (but still reasonable) reproduction scenarios where the difference *might* be audible, it would still only sound slightly different, not very different.
2a. Again, why? What benefit do you think you gain from keeping info that's inaudible?

3. As the name suggests, "Mastered for iTunes" (MFiT) means mastered specifically for AAC 256vbr. Assuming you download the original AAC 256vbr file, you are getting an exact bit perfect copy of what the mastering engineer released/intended. So, unless you have some serious fault with your equipment, if it "sounds funky" then it's supposed to sound funky and why would you want to change "funky" music into something else?

It's pretty clear that you don't really understand what you're looking at with a spectrogram and how it relates to either what your earphones are "trying to make" or what is actually audible, as others have effectively stated. Furthermore, as @pinnahertz stated, unless you are doing your "experimenting" using ABX testing (at reasonable playback levels), then all you are accomplishing is the creation of a "strong perception bias" and FALSE results, that completely void your experiments!

G

 

[castleofargh]

It's a spectrogram so the levels indicate spectral density. The more detailed spectrogram (bigger FFT), the more frequency points and the lower the spectral density, because the signal energy is divided to more frequency points. that's why "yellow" (or about -30 dB) is the loudest we see. The blue is down at around -90 dB, but you need to integrate all those blue points on the frequency line to get the signal level at say 15 kHz to 20 kHz. So if the FFT size is say 1024, the amount of positive frequency points is 512 meaning the frequency range 15-20 kHz has 512 * (20-15) kHz/22.05 kHz = 116 frequency points. If all of these are at spectral density level -90 dB, the signal level at frequency band 15-20 kHz is -90 dB + 10*log10 (116) = -69 dB. That's still very low level considering how insensitive hearing is at these frequencies! The frequency line here is linear (far from how human hearing works) and makes the blue stuff more dramatic it is for human ear. The upper half (11-22 kHz) is just one octave and on logarithmic scale ~1/10 of the picture (if 20 Hz - 22050 Hz plotted)!

You're right. I forget not to just take every value written at face value. In one form or another I've made that very mistake at least 3 times in the last year. Last time was on a RTA counting the stuff on the graph as noise level, even though the app would have given me the correct value so long as I bothered to click on a given frequency. So even when I got all the work done for me, even after reading the great stuff from Audio Precision dozens of time, and even after knowing that I make that mistake often, I still manage to fall for it from time to time.
Thank you for pointing it out. @bigshot too, even if it wasn't as detailed an explanation

 

[vatch]

It must not be forgotten that very few instruments even make much noise at all with any real power at or above 20khz; a 96khz audio frequency is absolutely unnecessary and will mostly be very high frequency noise with no musicality even though you can't hear it anyway.

The below chart, "THE FREQUENCY SPECTRUM, INSTRUMENT RANGES, AND EQ TIPS" lists the frequency ranges of many popular instruments for reference:

http://www.guitarbuilding.org/wp-content/uploads/2014/06/Instrument-Sound-EQ-Chart.pdf

 

[pinnahertz]

It must not be forgotten that very few instruments even make much noise at all with any real power at or above 20khz; a 96khz audio frequency is absolutely unnecessary and will mostly be very high frequency noise with no musicality even though you can't hear it anyway.

The below chart, "THE FREQUENCY SPECTRUM, INSTRUMENT RANGES, AND EQ TIPS" lists the frequency ranges of many popular instruments for reference:

http://www.guitarbuilding.org/wp-content/uploads/2014/06/Instrument-Sound-EQ-Chart.pdf

As much as I agree with your first statement, the attached chart is very misleading. It apparently charts only the sounds fundamental, not any harmonic content. It also implies that above the charted frequency no content exists at all, which is completely incorrect. If you want to prove this, just make a recording of your own voice and filter it with the most basic low pass filter at the charted cut-off frequency.

The harmonic content of a sound of any kind gives it its distinctive character. The point here is that no sounds, harmonics or fundamentals, can be heard above the upper limit of human hearing, with the corollary that human hearing response is not flat with a brick wall cut-off either, but displays a filter roll-off characteristic that starts lower into the "audible" range, and is affected by specific sound pressure and masking principles. But even with all of that said, 96kHz sampling with 48kHz pass band is not necessary to replicate the original signal.

 

[71 dB]

1 --- You're right.

2 --- I forget not to just take every value written at face value. In one form or another I've made that very mistake at least 3 times in the last year.

3 --- Last time was on a RTA counting the stuff on the graph as noise level, even though the app would have given me the correct value so long as I bothered to click on a given frequency. So even when I got all the work done for me, even after reading the great stuff from Audio Precision dozens of time, and even after knowing that I make that mistake often, I still manage to fall for it from time to time.

4 --- Thank you for pointing it out.

1 --- It's not often I get this, but thanks! I appreciate it. My years in the university weren't time wasted after all.

2 --- It's easy to make these mistakes. It's good to always ask yourself "does this make sense?" Vinyl enthusiasts have tried to demonstrate vinyl to have larger dynamic range than CD with this mistake: Because of the properties of spectral density and vinyl (RIAA curves etc.) the noise floor spectral density level drops easily well below -100 dB at higher frequencies and then people compare that to the CD dynamic range of about 90 dB or so without realising that for CD the same spectral density of noise floor (dither) drops insanely low and is possibly only higher at the highest frequencies if shaped dither is used (meaning at the lower frequencies the noise floor is even lower!)

3 --- Errare humanum est.

4 --- No problem.

 

[71 dB]

It must not be forgotten that very few instruments even make much noise at all with any real power at or above 20khz; a 96khz audio frequency is absolutely unnecessary and will mostly be very high frequency noise with no musicality even though you can't hear it anyway.

The below chart, "THE FREQUENCY SPECTRUM, INSTRUMENT RANGES, AND EQ TIPS" lists the frequency ranges of many popular instruments for reference:

http://www.guitarbuilding.org/wp-content/uploads/2014/06/Instrument-Sound-EQ-Chart.pdf

Years ago I read about a listening test were people evaluated how much in their opinion sound quality dropped when 44.1 kHz audio was downsampled to 22.05 kHz cutting everything above 10 kHz away and the conclusion was surprisingly little! Here it's about frequencies people definitely hear (10-16 kHz). That's why it's silly how much some people worry about totally inaudible frequencies above 20 kHz.

 

[vatch]

Yeah, I agree. It does simplify it a bit much and there are things going on in the upper range. The main thing really is that audio reproduction above 20khz and also below 20hz for that matter is unnecessary and highly likely detrimental as there is more computation and possibility of error in addition to increased high frequency electronically induced noise. In essence, most speakers can't reproduce it anyway.

 

[bigshot]

I don't think there's much in the way of harmonics above 10kHz that wouldn't be masked by lower frequencies. If you roll off everything above 10kHz, you can hear a slight difference, but it doesn't amount to a hill of beans in the grand scheme of things. But audiophiles are known to focus on the hills of beans and ignore the grand scheme.

 

[Vamp898]

To finally beat this topic to death.

I compared an 24bit/96KHz Orchestra recording (Song called 39 from the Tokyo Philharmony Orchestra from the Album Miku Symphony 2019) with the same recording in 16bit/44.1KHz

I inverted the 16bit/44.1Khz Song and mixed them together, the result was, of course, the difference.

What was the difference? At the point where the biggest difference was, there was an -57db loud(""loud"") sound at 22-45KHz

I was curious what this sound was so i lowered the frequency and there you go: It was white noise.

So if you choose for CD-Quality and not choose the Hi-Res Audio, you are missing out on -57db loud ultrasonic white noise.

I did the same with an 24bit/44.1KHz Electronic Song (Naomi from LukHash) and the loudest difference was an -86db loud white noise at 20KHz

Hope this makes you happy and this case is closed once and for all.

By the way, the difference from an OGG to an FLAC is way higher. If you do the same i mentioned above to an OGG at Quality 10 (comparable to an 400kbit/s MP3 if that would exist), you can here almost the whole song in the difference. Way quiter than the original recording of course and in way worse quality (sounds like an 64kbit/s MP3) but there is the whole song there.

The biggest difference, with the equalizer tuned to my personal taste, was the whole song but at -36db and bad quality. The difference changes drastically from song to song and, if you use an equalizer, there are parts of the songs where, when you know the difference, it could be heard if you turn the song loud enough. I played the MP3 and then enabled/disabled the difference and at some parts, especially in the low frequencies, there can be a difference. But chances are small anyone will ever recognize it comparing the songs side by side. I could only recognize it enabling the differences, i could not tell a difference when comparing the whole song.

Anyway I'd always go for FLAC because, at least for me, this test showed an big enough difference that i dont want to risk it. At this level of difference, chances for the decoder to ass things up are high imho and my Audio Player Supports 2TB MicroSD Cards and for the 128GB Card i have in there, i payed 13€ so whatever, who cares.

I totally agree with the initial poster of this thread. You will never hear the difference from an 16bit/44.1KHz to any higher quality recording, no matter what you do. Its an ultra sonic white noise, trust me, you'll never hear that and don't want it.

Last but not least, i noticed that the difference is bigger when you download a Song that is available in CD-Quality and Hi-Res Audio. The difference is still to small to be ever heard, but somehow if i downscale the song myself, the difference is smallet than buying a CD Quality Audio in the first place. But this is not always the case, I only found one album with that from what I checked so far.

Because the price difference for an 20€ album is roughly 3-5€, i still buy the hi-res audio songs just because im to afraid someone screwed up the downscaling, but this is the only reason. There is no technical difference for an human being.

 

[themysticmuse]

It's 03:17 am here in Sweden and I've been reading this thread for the last 3 hours. This is the best thing since sliced bread. Thanks OP, I've learnt a lot.

 

[bigshot]

I inverted the 16bit/44.1Khz Song and mixed them together, the result was, of course, the difference. What was the difference? At the point where the biggest difference was, there was an -57db loud(""loud"") sound at 22-45KHz I was curious what this sound was so i lowered the frequency and there you go: It was white noise.

You were probably blowing up the air conditioning room tone for the recording venue. CD noise floor is ridiculous overkill. You can't even record in a recording studio with a room tone below -95dB. You'd have to be in Carlsbad Caverns or an anechoic chamber or something. The room tone in your living room is probably higher than 40dB. So to play everything possible on a CD at an audible level, you would need to boost the level above that living room noise floor to 40+95=125dB. You're in the range of causing permanent hearing damage in a very short space of time there. Don't try this at home, kids!

 

[theaudiologist1]

Man this thread is still going after more than a decade. Wow!

Anyways, none of this matters because 1 bit DSD wipes the floor with PCM.

 

[chef8489]

Man this thread is still going after more than a decade. Wow!

Anyways, none of this matters because 1 bit DSD wipes the floor with PCM.

As you are in sound science how about you post scientific evidence to back up your claim. You do know that 99% of dsd was recorded in pcm and converted right. There are very few studios that have the equipment to do everything in dsd.