I had the opportunity to participate in a perceptual coding experiment run on Dave Moulton on behalf of Lucent Technologies. The experiment was to determine the audible of 5 different perceptual coding schemes. I of course was interested to see what else was out there, having only had experience with Dolby Digital and DTS (and some Sony Mini-disc, but not a lot).
Anyway, I went to Dave's house where he explained the experiment to me and another participant, and we started
The experiment consisted of listening to a ~10 second sample (the reference). He then played two more samples, one being the reference, the other being the encoded sample (A/B). The three samples were then repeated again. We were to determine which of the last two samples was the reference, and which was the coded sample. We ranked the A and B samples from a scale of 1 to 5, with 5 being "indistinguishable" from the reference, and 1 being annoyingly and distinctly different from the reference.
We listened to about 60 samples at a time, with a ten minute break between samples. We listened to the samples through headphones in Dave's living room/mixing studio.
We listened to samples from the following recordings:
I assume that every recording sample was played through every codec, although I have no verification of this. For some samples, the difference between the reference and the coded samples were definitely easy to spot. In particular, the castanets sample was one where I could pretty much spot the two every time by paying attention to the reverb and the sharp transients. The drums in the Chicago tune and the cow bell & cymbals in "Love Shack" were also dead giveaways at times.
The two classical snippets were the hardest for me to identify, as well as the spoken german text and "Nightfly" (except for the really bad codec(s) as you'll see below). However I'm not sure if this is due to the quality of the codecs or my unfamiliarity with the material.
Two things I noticed about the test:
Overall, I was very impressed. Assuming of course that I was able to distinguish between the codecs, there was one that was horrible, one which sounded different only when compared with the reference, and three that were very difficult, if not impossible to pick out. I was very pleasantly surprised, since I expected to walk in and easily identify the coded samples.
The one thing I meant to ask during the test, and forgot: the data rates of the different codecs. It could very well be the three codecs I like had a data rate approximately equal, or just below that of a normal CD. However if they were in fact fairly low, maybe perceptual coding has a future after all.
I have been informed by Dave Moulton that test results are on the way, so hopefully I will soon have an idea of how I did in the test
May 5, 1997
