Hi there, i don't want to start a command line / preset debat, just simple questions to asks :

1 - The switch -k (who disbale lowpass) is still in the 3.90.3 compile ?

2 - If i use it with -alt-preset extreme, did i have a chance to mess something or the only thing that the -k switch do is to add more bit to encode the frequencies who was striped by the lowpass ?

Soren

1. Yes.

2. You have a chance to mess something because bits will be used/wasted in a frequency range you're very unlikely to hear.

Hum, i'm a bit confuse if the file generated with the -k switch are bigger, i don't understand how some are wasted, assuming that wasted should mean that some bites who should go to the audible content goes to the unaudible ?? (IE here, the file size isn't a factor i want to discuss)

Soren

That is exactly what can happen.

It does seem paranoid, but mp3 is very inefficient above 16kHz, and it has a fixed 320kbps limit. So, when things get really though, wasting bits by encoding inaudible high frequencies could cause problems in the audible range.

If you encode problems samples, and observe the number of 320kbps frames (lame plots a nice little graph, or you can use EncSpot) then you'll see this isn't as paranoid as it at first appears.

Cheers,
David.

gday...



it can happen..
am using for mp3 -m s -k to extreme.. ( )
also use wavegain album before every encode..
the -k will in (to my ears) deffinetly improve the encoding
you also will shave off some bits.

but for adding -k to source with high gain/peak amplitude..
is a waste..

the bitrate increase is due to the (rather trivial) increase in frequency bandwidth.

as 2Bdecided says, mp3 is hopeless after 16k (aka in scalefactor band 21, sfb21 in shorthand). the mp3 specs don't actually define what is to be done with this area (i think most of the design was targeted toward samplerates of 32k).

sfb21 doesn't have a scalefactor of it's own - this is what defines quality in each band, by scaling up a band you increase it's signal-to-noise ratio, which is a vital part of the noise shaping process. in LAME, if the psy-model requires more precision in sfb21, the GLOBAL scale for the entire spectrum must be increased, leading to a huge bitrate penalty (scalefactors can only go so low, and their stepsize is large, so bands become limited to an unnecessarily high precision).

the -Y switch will disable noise-shaping in sfb21, so you'll save a lot of bitrate with this switch (but you'll lose a lot of the extra bandwidth that -k gives you). when -Y is on, the precision of sfb21 is at the mercy of the other bands, therefore more compressible music will have less high frequencies but a lower bitrate.

i myself use --alt-preset standard -Y, and haven't heard anything rotten going on for all the music i've encoded with it, so you might as well stick to that (average bitrate 180 rather than 200 for regular --alt-preset standard)

[edit]

anyone more knowledgable in mp3 please correct me if i've been mistaken in the above... i'm a layman, but a layman who reads a lot

Oh come on n68. You've been a HA.org member for a long enough time to know TOS#8 and know that such claims are utter nonsense (-m s -k). You still haven't delivered the promised samples/results regarding your claims concerning dithering and I doubt you will support your claim with facts this time.

If you don't change your attitude regarding this issue and take moderation advises more serious, you will be banned very soon.

ciao...


how long samples.. dev0..

You have to provide ABX results (original vs. --aps; orignal vs. <yoursettings>; --aps vs. <yoursettings>) and the sample you used to be taken seriously.
Otherwise your claims will be disregarded as totally bogus.

The info is in the answers above, but for clarification, the direct answer is : because MP3 is limited to 320 kbps. Thus sometimes, there is not enough room for every frequency, when more than 320 kbps are needed.
That's what "wasted" mean : a part of the 320 kbps is wasted for the upper frequency when lower frequency needs all the 320 kbps.