Will high frequency boost(?) be fixed?, So I can listen on my iRiver for xmas! Options

[listen]

Hey,
I think it's awesome that vorbis is getting support on some (more common) hardware. I think it would be a great format for blasting low bitrate tunes out in the car because it sounds surprisingly pleasant mostly. Except for the high frequencies, they are horrible. Is it likely to improve in this area, or is this not really considered a big problem? Should I just slap on a low-pass and enjoy my nice sounding (but muffled) music?

[c_haese]

Yes, as I report in http://www.vorbis.com/ot/20030930.html, the high frequency boost is considered a problem, and Monty has ideas for fixing it with Vorbis 1.1.

Of course, no release date has been set, so please be patient.

Best regards,

Carsten Haese

[ErikS]

Of course it will be fixed by christmas.

... Oh, you meant christmas this year? Nah...

[QuantumKnot]

Yes, as I report in http://www.vorbis.com/ot/20030930.html, the high frequency boost is considered a problem, and Monty has ideas for fixing it with Vorbis 1.1.

Of course, no release date has been set, so please be patient.

Best regards,

Carsten Haese

Just on a side issue, is the release of Vorbis 1.0.1 going to be delayed until the win32 build issue is resolved? Also, any idea on what sort of things we're gonna expect?

This post has been edited by QuantumKnot: Oct 3 2003, 16:29

[john33]

Just on a side issue, is the release of Vorbis 1.0.1 going to be delayed until the win32 build issue is resolved?

Yes, and the issues are being resolved as a matter of some urgency.

Also, any idea on what sort of things we're gonna expect?

Do you mean 1.1, or 1.0.1? If it's the latter, they're already in the CVS and incorporated in my most recent builds. If you mean 1.1, over to Carsten.

[QuantumKnot]

Also, any idea on what sort of things we're gonna expect?

Do you mean 1.1, or 1.0.1? If it's the latter, they're already in the CVS and incorporated in my most recent builds. If you mean 1.1, over to Carsten.

Well, as a matter of fact, I'm not even certain which one I was referring to.

I guess its 1.0.1. I mean, is it gonna sound a bit better than the previous builds?

[c_haese]

Well, as a matter of fact, I'm not even certain which one I was referring to.

I guess its 1.0.1.  I mean, is it gonna sound a bit better than the previous builds?

1.0.1 will sound a little bit better on certain files, but the improvements are minimal. One fix corrects the problem that causes soft signals to disintegrate into a twinkly digital murk. Another fix slightly reels back the midrange noise bias, and I am literally quoting Monty's CVS log message here, so I have no clue what exactly this means and in which circumstances it produces audible improvements.

Major improvements, including a fix for the HF boost, will come with 1.1.

Hope this helps,

Carsten Haese.

[double_chocolate]

I would just like to add that this high frequency boost introduced by Ogg Vorbis is noticeable through a wide range of encoding quality (e.g. 2-6) and is really mostly noticeable with metallic sounds with lots of high frequency partials, like cymbals. All the tests I've done show the OV trounces MP3 in spectral fidelity and overall sound and this 'brightening' (seemingly in the 14-16KHz range) is really unfortunate and should be a high priority. Cheers.

Pat

[QuantumKnot]

I would just like to add that this high frequency boost introduced by Ogg Vorbis is noticeable through a wide range of encoding quality (e.g. 2-6) and is really mostly noticeable with metallic sounds with lots of high frequency partials, like cymbals.  All the tests I've done show the OV trounces MP3 in spectral fidelity and overall sound and this 'brightening' (seemingly in the 14-16KHz range) is really unfortunate and should be a high priority.  Cheers.

Pat

You might want to try QKTune beta 3.2 which reduces the high frequency boost and pre-echo:

http://www.hydrogenaudio.org/forums/index....ndpost&p=187669

Another Vorbis encoder which has had success with HF boost is nyaochi's Modest Tuning beta 2:

http://homepage3.nifty.com/nyaochi/soft/dist/oggencmtb2.zip

Aoyumi's aoTuV Vorbis encoder also has had success with reducing HF boost:

http://homepage3.nifty.com/nyaochi/soft/di...nc_aoTuV_b1.zip

Try these and use the one which you think does the best job at reducing high frequency boost. All of them have been shown in various listening tests to be better than the official 1.0.1 encoder.

[music_man_mpc]

All the tests I've done show the OV trounces MP3 in spectral fidelity

That statment violates the HydrogenAudio Term of Service #8, I personally won't hold it against you as you are new here. Try comparing Vorbis to LAME 3.90.3 --alt-preset standard. I think you would be hard pressed to find a difference based on "spectral fidelity" as you call it. Also see this page for instructions on doing a fair comparison.

edit: Welcome to HA.org!

This post has been edited by music_man_mpc: Feb 28 2004, 17:43

[PoisonDan]

All the tests I've done show the OV trounces MP3 in spectral fidelity

That statment violates the HydrogenAudio Term of Service #8.

Well, I don't like the "spectral fidelity" part about this quote either, but I think we can safely conclude that Ogg Vorbis "beats" LAME MP3 at 128kbps in Roberto's 128kbps extension listening test:
http://www.rjamorim.com/test/128extension/results.html

I think you will also agree with me that Ogg Vorbis beats MP3 at low bitrates (e.g. 64kbps).

Of course, once we get to LAME --aps-like bitrates, things may be quite different, but it's hard to make a definite conclusion about it.

I just wanted to mention this, because I don't believe every post stating the superiority of Ogg Vorbis over MP3 is automatically a TOS#8 violation (although this particular post may very well be one).

[guruboolez]

I also agree that generally speaking, vorbis beats mp3 (lame and other). Nevertheless, if someone only listen for exemple to classical music, he could conclude on vorbis inferiority. I did some tests based on classical music only, and vorbis was never the winner - suffering from something "tonal purity": noise, coarse sound/details, etc... It's an old problem, and correction is just beginning.

[music_man_mpc]

[Off-Topic]

Well, I don't like the "spectral fidelity" part about this quote either, but I think we can safely conclude that Ogg Vorbis "beats" LAME MP3 at 128kbps in Roberto's 128kbps extension listening test:
http://www.rjamorim.com/test/128extension/results.html

I think you will also agree with me that Ogg Vorbis beats MP3 at low bitrates (e.g. 64kbps).

Of course, but double_chocolate made a blanket statment. Had he included "at medium to low bitrates" it would not have been a TOS#8, although I still may have grumbled about the choice of words (spectral fidelity), I probably wouldn't have posted.

I just wanted to mention this, because I don't believe every post stating the superiority of Ogg Vorbis over MP3 is automatically a TOS#8 violation (although this particular post may very well be one).

Agreed, although I definately believe that this is such a case .

@double_chocolate: I hope I'm not scaring you away, I made many such mistakes in my time of newbieness.

[/Off-Topic]

This post has been edited by music_man_mpc: Feb 28 2004, 22:06

[SebastianG]

for those of you who are interested in internals:

The "HF boost issue" is the result of 3 effects:
(1) low-SNR quantization with SNR maximizing thresholds (pretty constant 1 dB boost for SNRs around 5 dB)
(2) noise normalization feature (q<=5 increases energy only slightly IMO)
(3) point-stereo (q<=5, increases energy at worst 3 dB according to monty but strongly depends on phase correlations)

I'm not aware of the point-stereo code and how the manginute is calculated excactly.
But I do know something about the first 2 (they're related)

At the time I wrote that paper
voris dev mailing list posting + PDF paper attachment
I just focused on the first effect which remains the ONLY reason for q>=6.

I'm not that involved in what's monty up to now (who is?)
But let me tell you, that after a chat with him, I'm confident that he'll take this paper into account and fixes this issue.

More details:
Monty tried to "normalize" the noise energy by adjusting quantization threshold towards zero to compensate for negative QORs (see paper) but was not aware of the fact that the QOR ist mostly positive for SNRs above 2 dB. A simple solution would be to use the a similar noise normalizaton code to compensate for positive QORs as well.

bye,
Sebastian

This post has been edited by SebastianG: Mar 23 2004, 15:00

[double_chocolate]

[quote=QuantumKnot,Feb 28 2004, 12:29 AM] Pat [/QUOTE]
You might want to try QKTune beta 3.2 which reduces the high frequency boost and pre-echo:

http://www.hydrogenaudio.org/forums/index....ndpost&p=187669

Another Vorbis encoder which has had success with HF boost is nyaochi's Modest Tuning beta 2:

http://homepage3.nifty.com/nyaochi/soft/dist/oggencmtb2.zip

Aoyumi's aoTuV Vorbis encoder also has had success with reducing HF boost:

http://homepage3.nifty.com/nyaochi/soft/di...nc_aoTuV_b1.zip

Try these and use the one which you think does the best job at reducing high frequency boost. All of them have been shown in various listening tests to be better than the official 1.0.1 encoder. [/quote]


Thanks for the suggestions! Cheers.

Pat

[double_chocolate]

That statment violates the HydrogenAudio Term of Service #8, I personally won't hold it against you as you are new here.  Try comparing Vorbis to LAME 3.90.3 --alt-preset standard.  I think you would be hard pressed to find a difference based on "spectral fidelity" as you call it.  Also see this page for instructions on doing a fair comparison.

edit:  Welcome to HA.org! 

I apologize for the violation of site rules through my unqualified and rather bombastic statement. Thanks for the correction. Indeed, OV especially shines at lower bitrates and differences become less and less obvious beyond the 256Kbps range. The push behind my statement was also fueled by my disappointment with the Fraunhaufer performance. The differences at 96-160Kbps are not nearly as stricking when compared the LAME 3.90.3 encoder. I am working on a test set of complex sounds (e.g. various cymbal sounds) that I will record in studio and hope to post soon for any interested party to perform their own tests. Cheers.

[Aoyumi]

In order to improve this problem, it is aoTuV containing experimental hack. Please try.
Although some problems and the shortage of tuning are held, many problems will be improved simultaneously.
Supposing there is a big problem, please let me know. aoTuV test page

[harashin]

Seems to be improved.
Hustlejet Hosokawa___Atem_lied

[Aoyumi]

>>harashin
Thank you for a test.
By the way, is replaygain (vorbisgain) etc. used in the test in a sample like hustlejet? Since this sample is clipped completely, if it remains as it is, it will become a surely tight sound.

[QuantumKnot]

Again, 1.0.1 is the worst. Very consistent trend I think

[brett]

can someone explain why guru's statement ("i did some tests . . . vorbis was never the winner . . . suffering from something "tonal purity": noise, coarse sound/details, etc") is not also a violation of 8?? i am quite confused, tho admittedly slow . . . nothing personal guru -- love reading your posts

This post has been edited by brett: Mar 30 2004, 02:40

[rjamorim]

can someone explain why guru's statement ("i did some tests . . . vorbis was never the winner . . . suffering from something "tonal purity": noise, coarse sound/details, etc") is not also a violation of 8?? i am quite confused, tho admittedly slow . . . nothing personal guru -- love reading your posts

All posts by Guru understate he did a listening test to back that up. He is moved by listening tests :B

[guruboolez]

brett > do some search, please... I published tests for ~40 samples.

http://www.hydrogenaudio.org/forums/index....howtopic=16395&
http://www.hydrogenaudio.org/show.php/showtopic/14091
http://www.hydrogenaudio.org/show.php/showtopic/18359

[harashin]

>>harashin
Thank you for a test.
By the way, is replaygain (vorbisgain) etc. used in the test in a sample like hustlejet? Since this sample is clipped completely, if it remains as it is, it will become a surely tight sound.

OK, I dig it. This time I preferred QKTune 3.2 at -q4, gives the highest average bitrate of the four, on the same sample.
Hustle_Jet ReplayGained

[Aoyumi]

OK, I dig it. This time I preferred QKTune 3.2 at -q4, gives the highest average bitrate of the four, on the same sample.

Indeed, this may be the difference of the measure against preecho.
However, since the technique of solution of a HF Boost problem is different in QKTune 3.2, assertion is impossible.

Although there was no direct relation, it reconfirmed with this sample that there were many problems which originate in the circumference of a stereo in q4.