Welcome Guest ( Log In | Register )

LAME + Dibrom's presets/switches + remarks
post Jan 5 2002, 05:19
Post #1

Group: Members
Posts: 1394
Joined: 20-December 01
From: Bellingham, WA
Member No.: 693

ok...i have read and read and read trying to learn the technical details of the "--alt-presets" but i still have some questions:

1. what are the current switches in each of the presets [i.e. extreme, standard, and their "fast" counterparts].

2. could someone explain what the non-obvious switches do for those that aren't involved in programming/developing LAME

i think it would be a good idea, and probably save some newbie questions [not unlike this post, *tears*] to sticky all the information conceivably possible in regards to the "--alt-presets" and why these switches are used. i don't desire "the history of the presets" just what they do in the current/recommended compile. also, differences in the "fast" and "non-fast" presets [i.e fast standard vs standard] would be GREATLY appreciated [why? because, well...they seem pretty much the same to me audibly]

3. and could somebody explain to me the reason for writing or not writing the Xing Header?

4. what is a good program for analysing mp3's [i.e. to see the differences between encoding switches and the quality they produce within the same sample] that is exhaustive...

5. can anyone REALLY hear a difference between a CD track and their LAME "alt preset standard/extreme" counterpart?

ok...now for some remarks

1. this forum is awesome, i am learning just a phenomenal amount of information in minutes/hours/etc. thanks for this resource and don't let the "powers that be" get you down...[i.e. the official developers of LAME]

2. if LAME is open-source, and their are quite alot of people creating compiles, why is there still an "official" version...why not just release versions like "3.90dm1.1" and "3.90mitiok1.1" etc...this would clear up alot of confusion/egos/etc. and allow those that use certain aspects of LAME [i.e. dibrom's presets in dibrom's compiles] for those that desire them without all this feminine, drama...[damn flaccid ninnies]

3. to dibrom [and other devs w/ PM]: don't let this opposition worry you. anyone who is even *slightly* knowledgeable in regards to encoding/LAME/mp3 would use this site as their primary resource for EVERYTHING [compiles/information/forum]...i am sure that the opposition derives from lowly emotions, "if you know what i mean"

4. in no way shape or form do i think that releasing 3.90.2 was in bad taste, disrespectful, unbecoming, etc. and all who think so have some sort of issues...as far as i am concerned compiles from this forum ARE the official versions and i use and inform others only of this site.

5. man, 700+ posts in a couple months...geesh Dibrom...why do i have such a comfortable feeling that this post will be responded to quickly...thanx, again...

i don't really expect a complete response to all of this...it was more to just post my questions/comments. but responses to the questions would be greatly appreciated

and just a side note, it sucks having to "un-learn" everything from r3mix.net...man does that guy has issues...he seems like some sort of deluded fool after coming to this site, oh well...

issues, issues, issues, issues, and more issues...geesh

[btw...i have been reading these forums for about a month now, but this is my first post and i just registered not too long ago...i hope my questions seem more than retarded...just slightly more...]
Go to the top of the page
+Quote Post
Start new topic
post Jan 9 2002, 22:11
Post #2

Group: Developer
Posts: 2797
Joined: 22-September 01
Member No.: 6

Originally posted by xmixahlx but what is the advantage of then using a filter which chops off frequencies, such as the lowpass? is it a large difference in size? bitrate? or just that MOST people won't be able to hear these frequencies?
Well, it's very common to hear over 19khz sine wave in hearing test, but it's totally different to be able to distinguish 19Khz lowpassed music from original. So with vbr presets it's used to reduce the bitrate/size.
At lower bitrates - especially with cbr - using lowpass may be essential so that lower frequencies (more important for the quality) can get enough bits.
regarding the sgt. pepper dog whistle sample [again, sorry]:
is this a situation where the lowpass failed to produce an acceptable encoding? or was it the fault of another switch?
No, this is not a lowpass issue. Both fast standard and standard use the same lowpass, but fast sounds better here (which is certainly an exception).

regarding experimental switches: i ask this because on many of the forums which i have visited, there are countless many who swear by thier ridiculously exhaustive parameters in the commandline, and i have seen people swearing by their "--alt-preset extreme -X -Z" or whatever but it seemed to me that if these switches were so good, they would be included in the original parameters of the alt-presets
Well, alt-preset vbr modes use different -X modes and -Z switch more intelligently than was possible before. Before this, these were either enabled or disabled (or only one -X mode like -X3 was in use). Now the use is actually dynamically controlled depending on the need.
To fully understand what -X modes and -Z actually do in practise, you need to know something about mp3's quantization process in general.

-X modes are used to search for good quantization (the least audible noise). Different X modes define which criterions are used for that procedure.

-Z is a switch which is used to change between "noise shaping type 1" and "noise shaping type 2". When type 2 is used, it leads often to more use of scalefac_scale. When scalefac_scale is used, the quantization noise allocation (noise shaping) is more dynamic (*), and because of that, it can lead to audible artifacts in some special cases. The reason why type 2 is used is of course to reduce the needed bitrate. Dibrom's alt-preset vbr modes are changing adaptively between type1 and type2 depending on the situation.

MDCT coefficients are grouped into scalefactor bands aka SFBs. SFBs (21 for long blocks) cover the full frequency range and follow the spreading of human hearing critical bands. Quantization noise for each SFB is controlled with quantization stepsizes->the larger the stepsize of given SFB the more quantization noise for given SFB. Stepsizes are controlled with scalefactors. Scalefactors are used to reduce the stepsizes. There is one scalefactor for each SFB, except for the last SFB there isn't, which is an issue (not good) of MP3 standard..

When "noise shaping type 2" is used, it happens more often that scalefactors are multiplied by 2 (scalefac_scale), thus leading to increase of dynamics of stepsize reduction control. As much quantization noise is introduced per SFB as is allowed by masking for a particular SFB. When a scalefactor (which controls the reduction of quantization stepsize) is multiplied by 2, it means that there's more room/dynamics to control the stepsize value reduction, which may lead (in rare cases) to situations where stepsize is not small enough because the masking threshold may not be precisely calculated and/or noise quality measurement (-X) thinks quantization is good enough or bits just run out.

Noise shaping and quantization loop tries to find the lowest audible noise depending also on -X modes which define the criterion for "good enough" quality, which means the search for best quantization (hopefully least amount of audible quantization noise) ends for given SFB MDCT coefficient group and next quantization loop for next group of samples takes place. The criterions are listed at the bottom of this page:

Juha Laaksonheimo
Go to the top of the page
+Quote Post

Posts in this topic

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:


RSS Lo-Fi Version Time is now: 2nd December 2015 - 03:02