Slashdot Mirror
Kenwood Tries To Improve MP3 Sound
Wister285 writes: "Although MP3 quality is pretty damn good, the people over a Kenwood thought that it still doesn't have the edge that CDs do. MP3s don't support high frequency that regular CDs do because of the data compression. Kenwood's format, which is called 'Supreme Drive' (another dumb name for a good product ...), is boasting good results. Catch the story over at Excite." While it's cool that research is going on to improve the quality of compressed audio, it's hard to tell from this article just what is actually going on. Does it even make sense to say that this program "takes the missing harmonics -- known as 'fundamental' -- and mathematically re-processes the
data through a sound generator" to achieve a more natural sound? Where does it 'take' that information from exactly?
Aiwa's got one for around $300. And with that Kenwood (Z919 I believe?), it isn't like you're buying a $200 cd receiver and paying $450 for the mp3 capability. The Z919 is, from what I hear and see at least, a damn fine in-dash CD receiver. $650 does seem a bit steep though..I could pick up a good Sony Mobile ES receiver, a decent 4-way amplifier (75-100 watts), and maybe a pair of Infinity Kappa 6x9's for my back deck for all that :)
When a not is played, you get a pitch which corresponds to the name of the note, called the fundamental. Because of the acoustic characteristics of the instrument, you also get a bunch of overtones, which are pitches higher than the note in intervals such as fourths, fifths, and octaves to the fundamental. Different instruments produce different overtones, which causes its characteristic timbre.
Now I'm not entirely sure about the terminology I used above, but I think part of the point is that if it's going though and adding overtones, you aren't going to get a very natural sound, because everything is go to sound more similar. It might sound lusher, but it won't sound exactly right.
There's also the problem with a lot of music such as sacred music, which frequently employs high vocals, especially "castrati", now usually counter-tenors (men singing high), or boys who haven't hit puberty. I've converted some of that music to MP3, and although the high stuff sounds thinner than on CD, I don't think I want Kenwood lushening that sound -- part of the beauty of those voices is their purity.
Actually, what has a greater effect on the way something sounds is the attack. MP3 already handles this quite well for all but the most demanding tasks, when the amplitude of a sound is too high to fit in the selected bitrate and modulates ("shoves aside") the other material. Stravinsky probably sounds poor in anything below 256.
Anyway, what Kenwood seems to be doing here is our good old friend "interpolation". They've developed an algorithm like that used to enlarge photographs and applied it to sound. In order for this to work, they're going to have to create a device which can actually identify different instruments and supply the missing harmonics. The initial results, like when engineers attempted to create "stereo" from old mono recordings by channel equalization, is likely to be flawed, but I'm sure it will result in a commercially acceptable result. As for me, I'll be listening to Super CD or DVD Audio, whichever wins. I don't want Miles Davis' trumpet to sound like Maynard Ferguson's.
Gamingmuseum.com: Give your 3D accelerator a rest.
I actually own one of their older units. It has three buttons, a knob, and a power switch. To use it, you send a signal through it, twist the knob until things sound bright and shiney, and back down a bit. Sound non-technical? It is.
There isn't much to be seen *inside* the box, either. Aside from stuff which is standard fare in just about all audio processing equipment (trim pots, a couple op-amps, power supply, some relays...), there's only two devices which stand out. These are really large-looking devices in a bastardized DIP package, emblazoned with the BBE logo. Just looking at them, they seem to radiate magic.
But this is all off-topic, and pointless unless I give some subjective evaluation of what the magic does for music.
So, here goes. The effect on music is that it tends to sound a little livelier. Cymbols tend to have a little more detail, snares tend to jump out a little more. Bass sounds fatter, with more percieved string noise. It seems to have very little effect on a clean electric guitar, but can make a distorted guitar almost overbearing.
The effects are dynamic, and this can be heard when listening to a slightly noisey FM radio station. The noise will tend to breath (get louder/softer, and/or change in character) along with the dynamics of what's going on. This is most noticable (and annoying, once it is noticed) on spoken word.
That all said, I use it somewhat frequently. I've got a number of recordings which seem to lack life, and the BBE seems to provide some (even if it's a creative, or even destructive, process instead of restorative). It also does a bang-up job of fixing vocals that are turned to mud by a poor PA system, in a live enviroment, and has some usefulness in the studio.
I tend not to use it on MP3s that are heavily artifacted, as it just tends to enhance the artifacts more than the missing high-frequency components.
Given the apparent lack of details about the Kenwood Supreme Drive thing, one can only be lead to assume that they use a similar process to the BBE to "restore" (ie, create) lost harmonic information. If so, it'll be a useful thing. But, it will not be all things to all people, and no signal processor (no matter how many buzzwords you associate it with) will ever be.
(as an aside: Most FM radio stations process everything until it's just gelatinous muck, lacking absolutely any dynamic content, and with the spectral content smoothed out so that no song sounds and better or worse than any other - instead, they all sound bad. The MBAs who play general manager say it's good thing, because a) it makes their signal as loud as (or louder than) the competitor across town and b) they think the consistantly-mediocre quality will entice listeners to stick around longer than they would if they could hear the true nature of a recording. Frankly, it just makes me flip the dial to NPR or one of the local college stations, as they suffer from none of the hideous all-things-to-all-people processing that the 50,000-Watt buggers do.
It's unfortunate that people feel the need to have "digital" radio, when standard analog FM could sound almost perfect (and certainly better than MP3) if they'd just stop fucking with it.).
Kid-proof tablet..
Take a look at this link:
http://www.r3mix.net/
Basically the author encodes various songs and test
The results show a range of differences between encoders. The most popular encoders (xing), which I had been using myself in Music Match (latest version replaced it w/ Fraunhofer I think), just whack off all the frequencies above 16K hertz, no matter how high the encoding bitrate. As I understand it, that is just an arbitrary decision made by whoever implemented the encoder. If you encoder goes to 22Khz, is the Kenwood technique really necessary?
Another very interesting surprise was the finding of a bug in the latest Fraunhofer encoder as used by MusicMatch. Using the "Very High" quality setting (most people don't - it drops encoding to about 0.2x speed) the results were much worse than low bit rates at lower qualty.
What we have is no real consistancy in MP3 encoding between different sources. Different people use different rippers, encoders, and bitrates. I can download the same song 5 different times from the net, and I'll bet the files won't be identical.
And therein lies the problem as I see it -- this processing approach that Kenwood is working is on is going to vary in effectivness depeding on the encoding of the MP3 in question.
One thing I wish was done, was for there to be fields in the MP3 ID3 header for:
1) Encoder/Software Name
2) Encoder/Codec version #
3) Encoder setting (Bitrate + options)
These would be a great use in determining quality at a glance, as bitrate alone doesn't tell me that much.
The truth is most people I know use MusicMatch at about 160Kbps on Fast mode. It's a matter of convience - being able to just stick the CD in the drive, and in 10 minutes it's ripped using digital extraction from the CD. I've done it myself on about 1600 songs from my personal collection of 1200 cd's.
Now, I'm reaching the point where I can tell the difference between 160k and 192K on some songs. Add to that what I've learned about encoding (and the fact that it's constantly evolviong) and the questions of what to do arises. Should I re-rip my CD's with LAME? (I did that once to go from 128K to 160K and MusicMatch). Should I Wait? Should I just go on?
And finally, I still have this question: What about playback? Is there any difference between playback engines? I've got a RIO 300 w/64 Mb and I use it all the time. If I buy a newer device will it sound better? When I did a JPEG decoder years ago, I put in two options for the IDCT - faster vs higher quality. Not much difference between the two, but some.
They cite the possible uses as car systems and portable audio, but these aren't perfect listening environments. In either situation, background noise is going to effect the sound quality anyway. They don't tell what bitrate they're improving the quality at, either. This may just be a way of trying to fill in the gaps at 128kbps or some other low bitrate, where a higher bitrate MP3 would still sound better.
Brilliant eh?
Where does it 'take' that information from exactly? Having studied harmonics and data compression, and played with mixing both, I can tell you exactly where it comes from. I've narrowed it down to a few possible sources:
1) int(rnd * 255) + 1
2) "drop your shorts, bend over... this is going to feel a bit snug, but we have to get those missing harmonics."
3) in case you didn't pick it up that was a reference to OUT OF THEIR ASS!
4) (telephone ringing) "hello?" "yes, we were just wondering if you've seen any missing harmonics recently, or if you have any you could donate." "Oh, sure! I've got a box of those in my garage!" "Thanks! We'll send somebody over to pick them up. Please leave them in a box outside of your house."
5) int random_harmonics(); (sorry if this doesn't look right, it's been 3 years since i've coded c)
6) maybe they're LYING! do you honestly think that you'd be able to tell the difference?
"I hope I don't make a mistake and manage to remain a virgin." - Britney Spears
Does this new Supreme Drive work on other sound formats? If not, Ogg/Vorbis is on its way - maybe Kenwood should focus on that instead.
;-)
Is this new tech actually just guessing the new high frequencies based on the sound it "hears"? If so, that's adding to the music in ways that might not actually work. And this has been done already - see Wowthing, which although being pretty cool, can murder some songs (I'm thinking Bon Jovi, here
People who really really really *really* care about enhanced quality are probably going to buy the original CDs anyway, and won't be interested in buying (I'm assuming buying) Kenwood's Drive.
MP3 is still proprietry. This is not a good state of affairs. Kenwood developing for this is not what I want to see =)
--Remove SPAM from my address to mail me
My guess is that the software computes a spectrum (using short-time FT or Wavelets, or some other method), looks for harmonic patterns in the lower frequencies (which tend to be attenuated less by lossy compression techniques), and thus regenerates high frequency data to fill in the attenuated harmonics. Audiophiles are probably NOT going to like the results (as well as being philosphically opposed; by its nature it trades one type of harmonic distortion for another), and non-audiophiles will be mostly indifferent, IMHO.
Still, I'd like to listen to the results on some good monitors...
"It's overkill, of course. But you can never have too much overkill." - Anonymous Slashdot Coward
I'm afraid the only way to deal with this is to be harsh. It's utter rubbish. The whole point of lossy compression is that it removes elements bof the audio signal that are 'masked' by others, such that their absence is minimally noticeable. Once these elements are gone, they're gone for good.
Apart from the fact that the Excite article is embarassingly technically inaccurate, e.g. "Supreme Drive takes the missing harmonics -- known as 'fundamental'", it's obviously just a rehashed press release.
All they can do is add distortion - now that distorion may in fact have a 'natural' or pleasing sound to it, just ask anyone who prefers valve (vacuum tube in the US) hi-fi amplifiers, by virtue of being mostly even order, but it's distortion none the less.
Ugh, I hate technobabble, especially of the purposefully misleading kind. Anyone who understands the technology and claims this is meaningful is media whoring.
Told you it would be harsh.