--- Targets
* Preview:
should be low-bitrate enough not to be a bandwidth hog, but still recognizable not to blame the music basing on "this". Should rarely get saved.
Reality: MP3/CBR/24[22]
Ideal: ???
* Careless listener:
could be happy with "hey, it's playing", or
be constrained with the audio equipment available at hand -- which could eat up all the difference in quality, or bandwidth to use for background music.
Reality: MP3/CBR/112--128
Ideal: Ogg/64
* Careful listener:
well it could be rare one with decent ears, good phones and big pipe, but generally the one that has stored the record to enjoy it with both bright equipment and dim outside world.
He or she would be scared with clicks, distortions, flat emptiness across the beautiful track!
Reality: MP3/(mostly?)CBR/128--192
Ideal: MP3/VBR/192+ or Ogg/128+
--- Formats
So let's look how our Formats accomodate these Targets being used "from" given Media.
* MP3
Most importantly, it doesn't require any additional software to be installed or firmware upgraded in vast majority of cases.
What's worse, many commercial encoders suck.
One of the best (arguably *the* best) encoders is LAME, still the popularity of MP3 seems to cause low recognition of the importance of encoder and its options.
* Ogg
Not that hassle-free. While most free*nix systems provide neccesary libraries/plugins out-of-box, Windows and Mac users have to get the plugin (these are available for a stack of players though).
Its hugest plus is that a track driving at some given average bitrate is "standard" and best-sounding for that bitrate, no hassle this time and no frustration over otherwise decent music spoilt by crappy overhastened encoder.
=== practice
Well, I've changed my mind somewhat since my last listening test, but environment and software versions are quite different too (see below)
Given lame-3.93.1, vorbis-tools-1.0 and Blind Guardians--Surfin' USA (The Forgotten Tales) CD rip, I've done some testing aiming the targets.
It is, unfortunately, only "careless" environment at hand -- onboard sound"chip", so-so Philips phones (SBC HP140) and some Sven plastics (SPS-608 -- emulating Joe Average
, and finally a noisy Xeon server in the corner.
Original track was WAVed by cdparanoia, and downmixed with sox to have identical source for low-bitrate tests (just in case). Lame options were --preset cbr $BR -Z -h, the only ogg option was -b $BR.
24kbps versions were 22kHz/mono*, and lame did a surprisingly good job at that size: for preview it didn't matter with ogg (albeit blind listening shows ogg as a bit less deaf).
[*] doing less than 22kHz would kill off the remnants of treble, and doing stereo would be funny ;-]
128kbps (44/(j)stereo) was quite good; in fact, I couldn't guess lame's CBR/128 from original WAV for some 30 seconds from start (but remember the chip and the noise).
128kpbs (average) Ogg was no difference too; btw, oggenc without any hints made it into ~106kbps. (-v 3)
The thing is, that 64kbps (44 kHz) Ogg made it sound just as good as 128kbps MP3! So regarding "careless streaming", that would be the best bandwidth-wise.
My previous listening tests on lower-noise CMI8738 chip together with custom Philips-based amplifier and high-quality (not high-end, just Class 0 *) USSR acoustics) have shown that oggenc-1.0b4 was audibly better than lame-3.87 with 128kbps VBR, and that 128/144 kbps Oggs made no difference to me given the moderate, but non-zero noise levels of a city flat.
[*] Electronica 150AS-002M, three-band w/passive 'radiator'(?), 150W nominal rate, 320W -- musical power, 25--35000 Hz nominal freq range (+/-3dB)
=== conclusions
So while it may make no difference quality/bandwidth-ratio-wise to take on "24k preview" or "128k" classes (given that streaming MP3 were Lame ones with adequate options),
* "lo-fi" streaming with 32kbps/22kHz Ogg would
do something more reminding of the spectrum of original (but won't fit 33.6 modems well);
* streaming could get twice lighter without reducing the "careless" sound impression with Ogg;
* archival could get apx. twice lighter since it seems that people who use mp3/vbr/256 quite bear with ogg/128, and people who do higher don't see recognizable difference between ogg/160 and the original CD quality. Well, it could be my speculation but
The best applications of Ogg Vorbis at DMusic thus could be these two things.
PS: my apologies to Jagwac, seems that he enjoys best mp3 software like lame and rarely hears awful 128kbps from less advanced encoders... or did lame yet improve upon 3.8x (AFAIR) where I could still hear even artifacts?..
PPS: hard to explain, but even low-bitrate Oggs sound somewhat softer and less flat/deaf, higher-bitrate ones are also more full and don't have a superfluous "hiss/metal" in treble.
--
Michael Shigorin
16.06.2003
Regarding the thread I started at
http://www.dmusic.com/forum/complaints/4206:
the argument around 128/192+ mp3 is void since highly personal -- anyways, let's not fight over that. I've started with the suggestion to _extend_ (or allow for that), not to kill it, pluck in something new and see.
---
We have { targets, media, formats }
* Targets are roughly: { preview, careless listener, careful listener/archive }
* Media: { storage*, streaming }
* Formats, roughly: { mp3 { cbr, vbr }, ogg }
// [*] it can be streamed, pigeon-carried, but gets listened to from some
// solid storage, not the thin air
Targets require some minimal quality, which is in turn a function of Format and bitrate.
Everyone seems to agree that streaming is lighter with less bits to go.
Most seem to agree that MP3/128kbps (even lame's one) is not enough for archiving and further listening with even a half-decent audio subsystem. (even better _computer_ speakers could do a difference with WAV, not saying about better equipment).
Some know there are better formats than MP3, but it is still the most common denominator regarding both records available and software/firmware support, and the most recognized "brand" between music compression formats.
[ tbc ]
