WHAT FREQUENCIES ARE CRITICAL FOR REVERSALS?

By: George Duisman

David said that RealAudio tends to "zap" reversals. The practical upper limit that I've measured for RealAudio is about 3k Hz. That means that any frequencies higher than about 3000 cycles per second do not get transmitted via RealAudio. This is just about the same result as using a sample rate of 6000 samples per second. One can easily check out the frequencies in a sound file by using Cool Edit's frequency analysis. Any frequencies that are down from the peak level by 30 db or more are probably not heard. It would be interesting to find out what portion of reversal are "zapped" at the loss of any given frequency. This could be tested in the following manner:

1. Have the expert reversal finders store reversals for future study. These reversals should have been recorded on tape with good equipment or at 36,000 samples per second or better. A frequency analysis should be done to make sure that all frequencies up to 15k Hz are there.

2. Once we have 100 or so reversals like this, then have people with good hearing listen to these reversals, and note those that they can hear. Then using Goldwave's filter- high/low pass function, chop off the upper frequencies about 25% at a time until the reversal can't be heard. The sharpness of drop off should be max steepness. I would suggest that the frequencies used should be 11,000 8000 5500 4000 3000 2500 2000 1500 1100 800 600 & 450. The upper frequencies in this list correspond with common sample rates. The filter function should be undone (UNDO) after each filtering to avoid subtle effects of the filtering from accumulating.

3. Then a note is made for the last frequency range that allowed the reversal to be heard. This then would give us a good idea how many reversal would be lost on average by lowering sample rates etc. We would then graph this data for a visual picture of critical frequencies for reversals. It might be found, for instance, that about 3% of reversals are lost by using 22050 samples per second (11,000 Hz. drop off frequency), 5% more when the frequencies above 8,000 are cut off, but 30% when 6000 is the cut off. This would be very useful in choosing a sample rate.

It seems that reversals are quite complex and the above test would answers only some of the questions about reversals & frequency. But the reversal data base could be used for future research.

AUDIO FILE SAMPLE RATES

What happens to sound files at different sample rates?  Does the sample rate greatly affect the hearing of reversals?  Here is some preliminary results of testing one reversal "You're really a God" These graphs come from Cool Edit's spectrum analysis.  The first one is the original file.
 

 The main sound volume occurs at the -36db level.  Any sound frequency below -60 db is very faint.

The next graph shows what happens to the frequency graph when this 22050 file is converted to 6000 sample rate and then back to 22050.

The next graph shows the frequency spectrum of the same file transmitted via RealAudio and saved in 11025 format.

This last graph shows what happens to the frequency spectrum when this RealAudio file was converted to 6000 sample rate and then converted back to 22050.