Audio Formats Comparison here

[maikelnait]

I've done a detailed graphic quality comparison between many audio formats (Mp3 fhg, Lame 3.96.1, Lame 3.97a, Mp3pro, Wma, Mpc, Ogg, Aac and Vqf) at different bitrates.

The url is: http://www.telefonica.net/web2/audioformats/

I try to explain how the comparison is made on the web.
I hope it to be useful to you. I accept opinions.

Regards,
Pablo

[tedgo]

Very interesting comparison. But its not a good idea to recommend lame 3.97, cos its an alpha-version atm.
And your "graphical results" may be improper to suggests results about sound quality.
But i've made similar experiences in my abx-tests except the behaviour of mpc at about 256kbps...

[Brother John]

Hi, maikelnait.

Thanks for sharing your work with us all. I found some of the graphs quite interesting to look at.
However, if you wanted to find the best sounding codec, your test is hardly a reliable way to do it. And that's because it isn't suited to measure perceptible sound quality in a real world scenario. Let me explain why.

1. The source used is a totally artificial sound not reflecting real world material in any way. True, every codec had to deal with that same source file, so you may make valid statements about the codecs' relative performance (performance here meaning the ability to recreate the original spectrum as perfectly as possible). One should expect that for real life encodings those relative performances won't change too much. But you cannot be certain. An encoding technique suitable for the real could perform really crappy on artificial samples. Also codecs are typically optimised for real music. Just dig around a little on hydrogenaudio.org for the huge amount of testing that went into lame 3.90.3.

2. Spectrum analysis is not suited to draw conclusions regarding sound quality because spectrums only tell you how much mathematical deviation there is from the source. But lossy codecs don't even try to minimise that deviation. On the contrary. Virtually every lossy codec uses some kind of psychoacoustic model, which by nature lets the deviation grow. You said it yourself: frequencies can be masked or be too high/low to hear them anymore. A good lossy codec should drop those frequencies and spend available bitrate on those parts of the sound you actually can hear, thus enlarging deviation.

Of course, a perfect mathematical representation of the original means perfect sound quality. But if that's your goal, lossless is the way to go. Lossy codecs only can aim for transparency: no perceptible difference when listening.

What your test would need is an accompanying listening test. You might come to a conclusion like: Codec A clearly sounds the best, but it also has one of the greatest deviations from the source. As it's sound quality that matters, the graphs would have fooled you without a listening test.

A technical note: Why do you restrict all the codecs to CBR? E.g. Vorbis is a typical VBR codec and will perform significantly worse when restricted to CBR. (I didn't even know that any Vorbis implementation supported CBR yet. You sure you didn't get an ABR file? Still, ABR would be suboptimal.)

[BigDid]

Originally posted by maikelnait
I've done a detailed graphic quality comparison between many audio formats (Mp3 fhg, Lame 3.96.1, Lame 3.97a, Mp3pro, Wma, Mpc, Ogg, Aac and Vqf) at different bitrates...
Hi Pablo,

I have no opinions because I am no audio expert :)
; but I would very much appreciate if you could add AC3 audio format in your comparison at least at 192kbps & 256 kbps either with besweet and ac3enc (free) or softencode (commercial) or both because there seems to be a real quality difference between the two.
Tks

Did