[donunus]

I just posted this on hydrogen to ask their expert opinions about this thread... here it is... http://www.hydrogenaudio.org/forums/...howtopic=53707

 

[Publius]

Not much to say here, except that I agree with pretty much everything Febs has said.

Spectrogram analysis is only useful when properly interpreted. There are useful things one can intuit from them - note, in the vorbis spectrograms, how masking is done next to the dominant tone. This is a perfectly reasonable thing to do with a lossy encoder, and represents an advanced psychoacoustic model - and yet by Sir Nobax's reckoning, since it appears "not as accurate" to the original, it sounds worse. This is absolutely and factually incorrect.

Sir Nobax is doing nothing wrong or incorrect per se by running the spectrogram plots. Spectrogram sweeps are fun to look at and can indicate important things about encoders. But the conclusions he raises from them are deeply misleading and factually wrong. They will, at best, lead people to choose encoders that are needlessly bitrate-hungry to achieve some sort of numerical superiority that has no subjective rationale. At worse, it can lead people to choose numerical superiority over subjective performance, and people will wind up picking worse-sounding encoders.

In conclusion... look at the pretty pictures, then go to HydrogenAudio and base your encoder decisions from there.

 

[hYdrociTy]

Quote:

Originally Posted by Febs /img/forum/go_quote.gif Headphones do not intentionally discard inaudible data. Lossy codecs do. These graphs are nothing at all like a headphone's frequency response graph.

EXACTLY! I've been trying to vocalize that but obviously febs you are way more competent at these things due to your experience in battle

People keep saying well i dont care about pictures, i care about how it sounds.. well how the *&^% can it sound anything if it's not there? its either there on the graph or it's not, anything otherwise is just your imagination people...

 

[gorman]

Talk about visually beautiful and utterly useless at the same time.

Please, please, no one uses Blade over Lame based on these pictures, they are so far from telling the whole story it's ridiculous.

Nothing against the pictures in themselves, they're quite fascinating... it's just that they're not very useful for the intended purpose...

 

[kelvinz]

neat

 

[Sir Nobax]

Quote:

Originally Posted by Publius /img/forum/go_quote.gif This is a perfectly reasonable thing to do with a lossy encoder, and represents an advanced psychoacoustic model - and yet by Sir Nobax's reckoning, since it appears "not as accurate" to the original, it sounds worse. This is absolutely and factually incorrect.

Can't remember me saying that OGG is a piece of unaccurate crap, i've only said that the encoder forced VBR, making the file uncomparible with the others, wich average at the said bitrates.

But if a spectogram looks near perfect, and another looks flawed. However they both sound the same, while they are both the same bitrate. What technique is more accurate: spectogram's or ABX?

Because some people here say "The pictures suck, psychoacoustics know what you hear and what you dont, even if the picture looks like a 100% carbon copy the psychoacoustic model R0xx0r true compressions b0xxor".
Psychoacoutics arent perfect, I made this example earlier and ill do it again. The pychoacoutic model of a encoder cant know if the listener has a hearing loss at tones around the 4000hz. However the encoder simply thinks that the 4000hz tone overshadows the 1000hz tone, making the wrong dicision in this case as the listener doesnt hear the 4000hz tone, and the encoder deleted the 1000hz tone. This applies for everyone, as nobody has the same ears, but they all SEE the same. (except for a blind guy, but i dont think he would be reading this forum, would he?)

And if it LOOKS the same it ALWAYS SOUNDS the same, vice versa it doesnt.

 

[liquidfireboy]

artard.

 

[Gurra1980]

Quote:

Originally Posted by hYdrociTy /img/forum/go_quote.gif EXACTLY! I've been trying to vocalize that but obviously febs you are way more competent at these things due to your experience in battle People keep saying well i dont care about pictures, i care about how it sounds.. well how the *&^% can it sound anything if it's not there? its either there on the graph or it's not, anything otherwise is just your imagination people...

You are so correct!

I think it is important to have such graphs to look at, because if I lets say has only a portable player and a pair of cheap earpluggs there is no way I could hear a difference in higher bitrates, thats why I think it is important to have such graphs, some day you might end up with a pair of lets say UE10, that can detect small faults in the file. By looking at those graphs you get some help to choose the right encoder.

And to listen to what people on forums say about different encodings sounds like is just a waste of time because everyone says different and one has no idea what equipment they are using.

 

[sayrum]

hello.
i've been looking at this thread and trying to decipher the graphs + info here...
no luck so far (although i find they looking beautiful)

encoding my cds in lossy(?) is my wish.
was looking at ogg vorbis, heard that it supposedly has better sound quality then mp3 in smaller data packs.
after reading this thread i'm not so sure anymore.
so far i've listened to 192 bit mp3s (ripped with emule

) on my iaudio u3 - no amp!
i liked the sound quality.
although the few ogg files i have seemed to sound "more full" + still i felt something was missing..

i feel like a newbie: what codex can you guys recommend ?

 

[rb67]

I like to use mp3 V0 or 320kbps. If you own the CD's, you owe it to yourself to keep pristine versions. You aren't saving THAT much by switching to 256.

 

[Hanafuda]

this thread convinced me that starting over with AAC 192k VBR was not such a bad idea after all, even though its taking me forever. thanks for the great tests and reference info!

 

[Febs]

Quote:

Originally Posted by Sir Nobax /img/forum/go_quote.gif But if a spectogram looks near perfect, and another looks flawed. However they both sound the same, while they are both the same bitrate. What technique is more accurate: spectogram's or ABX?

One more time, since you still haven't gotten the point: ABX IS THE MORE ACCURATE WAY OF DETERMINING WHETHER THERE IS AN AUDIBLE DIFFERENCE.

Quote:

And if it LOOKS the same it ALWAYS SOUNDS the same, vice versa it doesnt.

You haven't tested this hypothesis, yet you just keep asserting it as if it were fact. Moreover, you haven't demonstrated at all that these graphs have anything to do with predicting or observing how these codecs perform with respect to music as opposed to pure tones.

When donunus posted on Hydrogen Audio about this method of evaluating codecs, one of the Hydrogen Audio members posted this graph to make the point that pictures do not always tell the story:

You should read the Hydrogen Audio thread, particularly the part about why the Blade-encoded graphs look "almost perfect."

 

[EnOYiN]

I was just going to post that picture here as well. You beat me to it Febs.

 

[Gurra1980]

Quote:

Originally Posted by Febs /img/forum/go_quote.gif One more time, since you still haven't gotten the point: ABX IS THE MORE ACCURATE WAY OF DETERMINING WHETHER THERE IS AN AUDIBLE DIFFERENCE. You haven't tested this hypothesis, yet you just keep asserting it as if it were fact. Moreover, you haven't demonstrated at all that these graphs have anything to do with predicting or observing how these codecs perform with respect to music as opposed to pure tones. When donunus posted on Hydrogen Audio about this method of evaluating codecs, one of the Hydrogen Audio members posted this graph to make the point that pictures do not always tell the story: You should read the Hydrogen Audio thread, particularly the part about why the Blade-encoded graphs look "almost perfect."

I don't think ABX is the best way! Because today I am sitting here with 192WMA, MF X-Can v3, HD650 and I can hear a difference that I couldn't hear whith my 595+iriver.

If I had seen the pictures before I could have made a better choise.

If I would be deaf and saw 10 pearsons playing guitar I could guess that there was music coming from the guitars, but if I also was blind I couldn't have guessed it.

I think everyone here knows that those pictures are not perfect, but you can't say that they don't help, because if they weren't showing the truth at all, then the picture of a 16kbps AAC file could look as good as the WAV, don't you think?

I don't think the thread on Hydrogen said anything that hasn't been said here.

 

[Febs]

Quote:

Originally Posted by Gurra1980 /img/forum/go_quote.gif I don't think ABX is the best way! Because today I am sitting here with 192WMA, MF X-Can v3, HD650 and I can hear a difference that I couldn't hear whith my 595+iriver.

So, what aspect of the graph was it that told you that you could hear a difference with your HD650 that you couldn't hear with your HD595?

Can you look at one of the graphs and tell me whether you would be able to hear a difference with, say, a pair of k701s?