Quote:
Came across this announcement at the Toronto Star of a new release by David Lanois that includes this great quote on his opinion on MP3:

Quote:
Listening to MP3s on earbuds is like watching Gone With the Wind on a cellphone. In time, compressed audio files may be a historical footnote, a technical aberration. But if that's what people want, I'm going to sell it to them.

Triggered by recent threads thaqt mentioned in passing the subject of the sound quality of lossy-compressed formats, I wrote an article these past few days comparing them. You can find it at www.stereophile.com/features/308mp3cd/.

And a tip of the hat to webmaster Jon Iverson for his beautifully appropriate Sergeant Pepper's illustration.

John Atkinson
Editor, Stereophile

Great article, John!

Very easy to understand and nicely laid out. Extremely interesting.

And the illustration is indeed clever and well done.

Query: What settings do you use to get the nice clean line when you look at the spectrum of a 1kHz tone, etc.?

It seems that whatever window smoothing I use(Hamming, Blackman, Hanning, triangle, etc.) - and whatever other settings in the various software-based spectrum analyzers that I have - I get more of a steep hill than a tight crisp line - even with a pure sine wave. (I do get a nice-low noise level of -125dB or lower).

The best is when I use Blackman-Harris with an FFT size of 65,536 (the highest I can get). FFT overlap doesn't appear to make any difference.

Then again, I probably am being to critical. When you zoom in, how many HZ wide are your peaks? (If this isn't too personal).

I've heard that ziping your jpeg's will screw em up but I had no idea.

Great article JA. More like this.

RG

Quote:
Great article, John!

Very easy to understand and nicely laid out. Extremely interesting.

Thank you. As you can imagine, I am tired of MP3s being proclaimed as being of "CD quality"

Quote:
Query: What settings do you use to get the nice clean line when you look at the spectrum of a 1kHz tone, etc.?

65,536-point FFT, Blackman-Harris window, 15 seconds of data analyzed, 0.67Hz resolution, using Adobe Audition. I could have got away with using a shorter FFT and less frequency resolution.

Quote:
It seems that whatever window smoothing I use(Hamming, Blackman, Hanning, triangle, etc.) - and whatever other settings in the various software-based spectrum analyzers that I have - I get more of a steep hill than a tight crisp line - even with a pure sine wave.

The skirts that can arise with a pure tone arise from the fact that the FFT sample is not continuous at its beginning and end. If that can be arranged, a rectangular window is the most accurate. If not a Blackman-Harris window does appear to give the best compromise re: bin leakage, etc.

If there is a discontinuity in the audio data in the sample -- even a slight change in level will do it -- this can also give rise to skirts around the spectral line representing a pure tone.

Quote:
The best is when I use Blackman-Harris with an FFT size of 65,536 (the highest I can get). FFT overlap doesn't appear to make any difference.

What program are you using? There may be some assumptions made by the author that are in operation. For example, I have found that the FFT engine in SoundForge always has more bin leakage than that in Audition, even when both are set to the same FFT paramaters. The very best window I have found for spectral analysis of data like this is the proprietary 7-point Prism window that is supplied with that company's DScope. I didn't have that system set-up when I did this work, however.

John Atkinson
Editor, Stereophile

Quote:
The skirts that can arise with a pure tone arise from the fact that the FFT sample is not continuous at its beginning and end. If that can be arranged, a rectangular window is the most accurate. If not a Blackman-Harris window does appear to give the best compromise re: bin leakage, etc.

If there is a discontinuity in the audio data in the sample -- even a slight change in level will do it -- this can also give rise to skirts around the spectral line representing a pure tone.

This helps a great deal. Following your reply I did some research (knowing a bit what to look for) and found that with Blackman-Harris one gets the widest bins but the least bin leakage. I understand better what is going on now.

Quote:
What program are you using? There may be some assumptions made by the author that are in operation. For example, I have found that the FFT engine in SoundForge always has more bin leakage than that in Audition, even when both are set to the same FFT parameters.

Interesting. I was indeed using SoundForge (latest version). I have a few other programs, such as Sonar, which also include spectrum analysis. I'll take a look at these also.

However, I now have the skirt of a 1kHz sine wave down to a maximum width of 6Hz. These settings work great. I can now easily see the relative strength of harmonics in an instrument's tone, etc. - exactly what I wanted.

Thank you!!

P.S. Sound is inherently fascinating, isn't it?

John,

I found your article well written and informative.

Footnote 1 has me puzzled a bit. Wav files are not compressed and to my knowledge do not contain any redundancy or error correction/detection bits. Are you referring to the way raw data is stored on the CD itself?

I think the whole processes of ripping, storing and playback could use a detailed investigation. There are many links in that chain that could have detrimental effects not the least of which is jitter. Here are just a couple of questions that one would like to have answers to: What is the kmixer doing to my bits? Or: When I send my bits out the digital port are they the same (exact) bits I have stored on my disk?

Stereophile is probably not the best forum for this. Or is it?

Thanks,
Scott

Quote:
I found your article well written and informative.

Thank you.

Quote:
Footnote 1 has me puzzled a bit. Wav files are not compressed and to my knowledge do not contain any redundancy or error correction/detection bits. Are you referring to the way raw data is stored on the CD itself?

No, I was musing aloud that in uncompressed data, each bit has less impact on, contributes less to, the recovered audio signal than with a lossy-compressed representation, the context being the general communication of data. Thus if a single bit is dropped streaming raw data, the result may well be inaudible. But dropping a single bit with a lossy compressed stream may well result in an entire data block being muted, which will be audible.

In practice, as was pointed out by several correspondents, transmission systems using lossy-compressed data, such a satellite video, use extremely powerful error correction to minimize the effect of such dropouts.

Quote:
I think the whole processes of ripping, storing and playback could use a detailed investigation. There are many links in that chain that could have detrimental effects not the least of which is jitter.

I agree.

Quote:
Here are just a couple of questions that one would like to have answers to: What is the kmixer doing to my bits?

Probably something unpredictable. Best to bypass kmixer.

Quote:
When I send my bits out the digital port are they the same (exact) bits I have stored on my disk?

Only if you use a soundcard that uses, for example, the ASIO protocol and have the PC's volume control set to do nothing and have all sound effects turnd off.

Quote:
Stereophile is probably not the best forum for this. Or is it?

I don't see why not.

John Atkinson
Editor, Stereophile