CHOOSING A SAMPLE RATE

To save disk space & transmission time.

By George Duziman

Choosing the best sample rate can save a lot of disk space as well as much data transmission time.  RealAudio, MP3, telephone, TV, radio and other media have limitations on the frequencies that are transmitted.  Using Cool Edit's "frequency analysis, one can easily determine what sample rate to use for storing and sending files.  Here is a frequency analysis of a sound track received via RealAudio.

This file was recorded at 11025 samples per second and then converted to 22050.  Because it was recorded at 11025 there are no frequencies above 5500.  You can also note that there is a sharp drop off at about 2600 Hz.  That is the result of being transmitted via RealAudio.  So this file could be stored at 6000 samples per second (3000 Hz. cutoff) with virtually no loss of the useful sound.  The cutoff frequency is one half of the sample rate.

To check a file yourself, in Cool Edit highlight a small section of sound with voice, approx. 5 seconds will do just fine.  Then choose "frequency analysis", then "Scan". 

Advanced Method:

Some sound tracks are tricky with high frequency noise.  These files will need to be tackled with a more advanced method.  With a little practice this method should take a just few minutes.  

This is sound from David's first 2000 TV interview.  It looks as though there are some frequencies above 5500 Hz. At first I thought that 22050 was the right sample rate, but on closer inspection I was not sure. 

Method:  You need to be familiar with editing one track only on a stereo file.  If you're not familiar with this, then play with editing stereo tracks in Cool Edit first.  

1. Choose a small section of sound (like 10 to 20 seconds or so) with some voice and a fairly large section of silence, i.e. just the background noise.  Something between 1/4 second and 2 seconds of silence (background noise) should work fine. Copy this this into the copy buffer.

2. Open a new instance of Cool Edit and paste this section of sound into it.  

3. Convert this to stereo at the same sample rate.

4. Choose a track for voice and use the other track for the background sound.  In this example I choose the left track for the voice and the right for background noise.

5. Copy the section of silence (background noise), into the copy buffer.

6. Silence out the whole right track. 

7. Paste in the background noise from the copy buffer to the right track.  Then mark the end of the background sound on the right track with the cursor and paste again.  Do this until the whole right track has the background sound. You can also use different sections of background sound for greater accuracy.

8. It should now look like this. 

9. You can now directly compare the background noise in the right track against the voice and noise in the left track. Highlight the whole file (10 to 20 seconds), both tracks and then choose "Frequency Analysis" and then "Scan". You will see something like this:

On this this one I choose to check "Linear View" as it gives a better view of the critical frequencies.  The purple shows the sound from the background noise, and the blue comes from the voice track.  Notice that the signal (voice track) is higher than the noise all the way out to about 10,500 Hz, but only in a minor way past 4300 Hz.. So from this graph, 22050 (11,000 cutoff frequency) is the correct sample rate only if you want the very best sound. Using 11025 sample rate (5512 Hz) or even 8000 (4000 Hz.) sample rate would be reasonable. 

Here is another "Frequency Scan" from the Aussie TV interview, using all of the large silence sections from the whole show.  This scan seems to show that 8000 sample rate would be good.  There are some faint frequencies above the noise between 4000 and 6500, and it's possible that one could hear them.  Notice that the noise has a fairly smooth curve, and the voice track has some "lumps". This won't always be the case as some noise could have a narrow frequency band.

You can also highlight small sections of sound and do a frequency analysis on just a small section. The silence section (background noise) can be critical.  Some sections of the silence could be more quite than others.  You may even take several different sections of background noise from the original file and try those. After a while you will know what sample rate to use for RealAudio, MP3, radio, TV etc. without testing.

Using a sample rate of 8000 instead of 22050 saves about 64 % of disk space. Transmission time would be reduced to less than 1 third.

 Using a sample rate of 6000 saves about 73% of the disk space that 22050 would use. Transmission would take about 1/4 the time.

Body goes here