I created my own test mp3 file by using the voice recorder. Watching the clock I whacked a stapler to the desk at as close to a 10 minute interval as possible. Likely well within a second. I then transferred the file to the PC and viewed it with Audacity.
Results:
The voice recorder has the two “whack” sounds at about 10 minutes and 0.4 seconds apart. This could be caused by imprecision of my testing, but a longer test over 28 minutes gives about 2.4 extra seconds. This would indicate that the clock for the voice recorder is running an itsy bitsy amount too fast - but this error this too small to be audible.
Using Audacity I cut out the extra time to make the two clicks “exactly” 10 minutes apart, and then I added a pure 440 hz tone for the bulk of the 10 minute recording. This is independent of any crystal or other oscillator - simply a mathematical equation. This file was exported as MP3, ogg and wav files and put on the Fuze.
When I play the MP3 file back on the PC I can use my Korg CA-30 Chromatic tuner to measure the frequency of the tone that was artificially generated with Audacity. My tuner identifies the tone as A440 with 0 cents deviation - a little fluctuation of 1-2 cents is caused by ambient noise.
When I play the file back on the Fuze pushing the earphones up to the mic on the Korg tuner I get a pitch of A440 with about -20 cents of deviation. This is very close to the timing error of about 1.1%. I also copied wav and ogg versions of the same file to the Fuze, and when I play them back they are also 20 cents low in pitch. The time it takes to play the MP3 file from one click to the other is 10 minutes 7 seconds, consistent with a little more than a 1% speed error.
Here are two links that discuss the human perception of pitch and the “cents” scale of frequency…
http://hyperphysics.phy-astr.gsu.edu/Hbase/Music/cents.html
http://en.wikipedia.org/wiki/Cent\_(music)
Both of these sources claim that the limit of human hearing is about 5 cents. But the second goes on to say that many music students cannot hear a 10 cent difference. I must admit - I cannot hear the difference in their sample files - though the tuner picks it right up. Wikipedia claims that most normal adults can hear 25 cents difference, which is pretty much what we are looking at for my player.
The funny thing is the voice recorder is not effected, and the real time clock appears to work right. This give me hope that there is a firmware fix that can repair the issue. On the other hand, mine might be broken. Does anyone want to try my test files. It is easiest if you are also a musician and have access to a crystal based tuner, but a stopwatch will work as well.
