Hi wschwartz and all,
Your question prompted me to dig out my hard copy of the study, a good thing since I discovered I made some errors in my summary that I posted here (strictly from memory. . .).
I don't know if it's available online through a link. Here are details: The Canadian portion of the study in which I participated was done by the Communications Research Centre (CCR)in Ottawa, Ontario, in 1992. "Highly discriminating" listeners were brought in from the CBC (Canadian Broadcasting Corporation), Audio Engineering Society, and specialty AV publications (at the time, I was editor of Sound&Vision Canada). Some editors of US-based magazines also came up to join the listening panel.
The other participants besides the CCR in Ottawa were the BBC research center in the U.K., the Communications Lab, Australia, and TDF, France.
I also got the data rates wrong. Some of the low-bit-rate codecs evaluated were at 240 kbits/second, and others at 360 kbits/second. They all have technical designations except for Dolby AC-2, so I don't know if AAC played a part. However, I seem to recall that one or more of the transparent codecs were developed at the Fraunhoffer Institute in Germany and perhaps AAC is derived from that codec. The best two of five were remarkably good at 240 kbits/second.
Other interesting facts from the report: the Suzanne Vega track was from A&M label, 395 136-2. We also used Dire Straits' "Ride Across the River" (I'd been using that track myself for some years in the NRC blind listening tests of loudspeakers). Additional tracks that were useful in distinguishing anomalies contained solo harpsichord, triangles, and castinets, the latter from a Telarc CD "Spanish Suite" by Ravel, Telarc CD-80171.
By the way, there was a 3-hour training period with up to 30 trials for all listeners before the formal tests began the following day. We had the choice of listening through Stax Lambda Pro headphones (electrostatic) or a a pair of tri-amped pro monitor speakers used with an active crossover and Bryston amplifiers. Generally, the headphone listening was more revealing of coding artifacts and distortions than loudspeaker listening.
In the blind listening, the impairment rating scale went from a top rating of 5.0 (Imperceptible) to 4.0 (Perceptible but not annoying) 3.0 (Slightly annoying) 2.0 (Annoying) and 1.0 (Very annoying).
The final report was called "CCIR Listening Tests: Basic Audio Quality of Distribution and Contribution Codecs". The authors were Ted Grusec and Louis Thibault. The report was submitted to the CCIR Task Group 10/2 (doc.24), Geneva, Switzerland, 10 November, 1992. The statistical analysis of the double-blind tests was very thorough.
The rating scale we used contains useful descriptors. Sometimes we did hear artifacts but they weren't annoying so we'd give that codec a 4.0 rating (perceptible but not annoying). The Suzanne Vega solo vocal was very revealing and I recall giving "Annoying" ratings to several of the codecs because of the edgy sibilance they imparted to her voice.
Regards,
Alan
