How bad is MP3 really?

[nb2]

I've heard some excellent guitar music from Napster done at modest mp3 bit rates. .

Yes, I was told that guitar easily sounds very good, and should not be used to check the quality of a system.
It is often excellent in movie theatres.

[dave2010]

Nicolas

One can only imagine what goes wrong sometimes. Currently listening to Puccini - Preludio Sinfonico and Capriccio Sinfonico played by Claudio Scimone and the National Orchestra of the Opera of Monte Carlo. Disgusting sound via Napster.

Yet it can't just be the music complexity, as a few minutes ago I listened to
Richard Chailly playing exactly the same pieces also via Napster with the Berlin Radio SO. Somewhat better sound, though still rather raucous in places. Presumably somebody wasn't doing a quality check for the Monte Carlo version, which is rough throughout, when they made the MP3.

The Berlin version isn't by any means good, but is relatively less disgusting.

[MartinT]

The dilemma is that CD may have a technically archaic encoding format, yet the best CDs played on a high end player can sound absolutely magnificent. There is still more to be learned about the effects of digital artifacts.

[nb2]

One can only imagine what goes wrong sometimes.

Right, it is very hard to understand where comes the difference from.
Is it the same mastering ?
How radios or Napster performed the encoding ?
I am convinced good 320 kbps streaming can be almost impossible to pick up from CD, but you don't have often such bitrate, and with low bitrate, that is another story ...

[Marco]

The dilemma is that CD may have a technically archaic encoding format, yet the best CDs played on a high end player can sound absolutely magnificent. There is still more to be learned about the effects of digital artifacts.

Hear, hear!

Your last sentence is almost certainly a truism.

Thing is, Martin, how many people do you think own a genuinely top-notch CDP these days and are therefore au fait with what CD is really capable of?

Most people these days use mass-produced players with plastic DVD-ROM mechanisms (or even worse, DVD players!), cheap PSUs, and sub-standard build quality, both externally and internally, so how they can hope to know what CD is genuinely capable of, sonically, by using such inferior machines is anyone's guess!

The fact is most people today haven't heard CD done properly, and thus have embraced computer audio either through desiring the convenience factor, wanting to be up-to-date with new technology, and/or because they've heard some of the bland, tinny sounding monstrosities currently on the market posing as CD players, and then compared those to music streaming via a computer and decent DAC, and found the latter infinitely more preferable, simply because the quality of their CDP benchmark was so poor.

Now don't get me wrong, computer audio done well is phenomenally good and can more often than not sonically outperform even the best CDPs one can buy, *but* in my opinion there exists a misconception these days that any half-decent music streaming system will automatically outperform ANY CD player, which when you've had the relevant experience of using a genuinely hi-end machine, and compared the difference between the two on many occasions, is quite patently bollocks.....!

<Rant over>

Marco.

[dave2010]

Nicolas

Right, it is very hard to understand where comes the difference from.
Is it the same mastering ?
How radios or Napster performed the encoding ?
I am convinced good 320 kbps streaming can be almost impossible to pick up from CD, but you don't have often such bitrate, and with low bitrate, that is another story ...

After years of listening I still don't fully understand this. Some of my earliest streaming audio listening was to Radio Sweden P2, at 96 kbps and probably using Real Audio codecs at the time. I was aware that it wasn't ultra high quality, but it wasn't bad either. Later I found stations such as Belgium Musiq3 - which was outputting at 128kbps mp3, and that sounded quite good. The BBC DAB broadcasts use Musicam (MP2) for their encoding, and Radio 3 is usually at 192 kbps. A few years ago they tried to reduce the rates to 160kbps, and many people noticed the quality loss. They excused themselves by saying they had new codecs, and that the quality should improve. They reverted back to 192kbps. There are internal BBC reports which suggest that it's been known for a long time that the bit rate for DAB should be 256 kbps (or higher if possible) for good quality, but like most broadcasters they have reduced their rates. My view is that 192 kbps MP2 is more or less equivalent to something between 128kbps and 160 kbps MP3, and that 128kbps MP2 (Radio 2, Radio 4 etc.) is more or less equivalent to 96 kbps MP3. ClassicFM is the only other UK radio station which uses a relatively (!!!) high bit rate of 160kbps using Joint Stereo, but they manage horrible sound quality anyway.

There does seem to be something in the quality of encoding - some encoders (perhaps with more carefully chosen parameters) do do better at low bit rates. Most do a lot better at high bit rates, and until recently I'd assumed that anything above 256 kbps should be good with any codec. However the recent post about Spotify and Ogg Vorbis has made me rethink this, and I need to investiigate that further.

There could be several reasons for poor quality compressed audio, which I hear sometimes from sources such as Napster.

1. The encoding is just not done carefully enough.
2. The original source has out of band frequency components which interfere with the encoding, and produce even more audible artefacts than those from a source without out of band components.
3. The "original" is in fact itself also in a compressed digitial format. The FAQ for Ogg Vorbis makes it very clear that because OV compresses in a different way from other methods, such as MP3, that compressing (say) an MP3 source to OV causes much greater loss, with consequent audible disasters. The BBC has also noted problems with transcoding music from one compression scheme to another.
4. The decoding resolution is 16 bits. [can happen sometimes ...] Interestingly MP3 can handle up to 24 bits and at different sampling rates. I think most players use floating point arithmetic. Playing back MP3s through a 16 bit D-A can, I suspect, give lower perceived quality - though this might only be a slight effect. I do have a hunch that it can take the edge off a "good" MP3 recording.
5. The bit rate is too low.

Although low bit rate compressed music can sound bad, and sometimes this is inevitable, I submit that this (low bit rate) is often not in fact the reason why it is terrible, as sometimes with care it is possible to do a decent compression at relatively low rates. Probably systems such as Napster don't have any really strong quality control procedures, so there is no attempt to optimise quality. If everything is put through the same encoder at a fixed low bit rate, then it is quite likely that the results will sometimes be horrible - as noted.

Note one thing, which I think has confused people. Most digital compression schemes these days cope with wide dynamic ranges for the source material. Some compression schemes, such as NICAM, address dynamic range explicitly by normalising, but this only gives a few dBs of improved signal to noise ratio. Discrete Cosine Transform methods, such as MP3 or other sub-band coding methods, handle wide dynamic ranges simply by what they do to the coefficients in each time window sample, so there should not really be a problem with loud and quiet sounds.

[nb2]

After years of listening I still don't fully understand this.

I finally came to the conclusion that mastering is far more important than anything else.

As you said, If you are not absolutely sure that you compare 2 different streaming (online or compressed file) from exactly the same original recording, in its same 16/44.1 mastering, you are comparing apples and oranges.

I have heard excellent live concerts on internet radio mp3 128 kbps stream.
We all now very bad quality commercial CDs.

If you are sure that the original source is exactly the same, then the way it is encoded is important.
Live encoding on the fly for live radio streaming surely requires excellent technicians and engineers (but can give astonishly good results).

Another problem, when you listen to a high quality 328 kbps stream, who knows if it is not playing some poor compressed file at low bitrate ?

[dave2010]

Nicolas

I finally came to the conclusion that mastering is far more important than anything else.

Not quite sure what you mean here. There are many CDs which have equivalent MP3s or other compressed versions, and in some cases (e.g some on Napster) the compressed version is diabolical.
Is that what you mean - ripping?

Of course original mastering is important, and if done badly faults can be hard to remove - though even that is sometimes possible. It's always good to start with a good original. From there further processing will normally make things worse - though not always.

Re MP3 you might be interested in this programme - http://www.bbc.co.uk/programmes/b00slls3 If you listen during the next 7 days you might discover why Suzanne Vega is appparently nicknamed "the mother of mp3."

[Codifus]

MP3 encoders, even at 320 kbps, can vary immensely in quality. That number, 320, just tells you is that the file was encoded at that rate, not really how good it is.

The difference in encoders is like 2 engines that produce 600 Hp. One engine is a 2.0 liter four banger with twin turbos, boosted to the high heavens. The other motor is a 7.0 liter V-12.

When the 4 banger is producing its maximum output you can barely have a conversation in the cabin, but the V-12 is strumming along bubbly smooth and quiet. You can have an intimate conversation with your significant other while sipping champagne

The best MP3 encoder I've ever heard was LAME.

CD