>Unfortunately for one lecture (from Myoshin Kelley), 40kbps wouldn't >do it; that characteristic low bitrate transform coder "swishing, >swirling" artifact was really present in it. So I kept increasing the >MP3 encoding rate, but found that I could not completely eliminate the >artifact no matter what bitrate I chose. > >Finally I went back to the original MD and listened to it carefully; I >was shocked! By now my ear had become attuned to what to listen for, >and I could actually hear it in the original! I had in my hands a >*voice recording*, from an ATRAC Type-R machine, that I could actually >hear something fishy with. It was subtle, and chiefly during the >sibilants (i.e. "s" sounds), but it was there. The thing was, on >initial listening the MD sounded fine, the effect only became >discernable once I knew what to listen for (see also my comment at the >end of http://www.minidisc.org/generations.html about becoming attuned >to artifacts). This "familiarity" effect is well-known in the audio research community. It is an important consideration, when designing a formal listening experiment, whether to use "expert" trained listeners, or whether to use a sample from the general public. Generally speaking, if you are trying to focus on one or two aspects of the sound then you would use expert listeners, but if you are designing a product or algorithm to be acceptable to a certain given percentage of the population then you use a sample of listeners from that population. This use of "population" is in the statistical sense - if your product is intended for use solely by accountants then you use a sample of accountants to judge your design. A friend of mine had his wedding recorded on MD by a "professional" sound engineer (actually a professional photographer who owned a couple of microphones and an MD and charged extra to bring them along as well as the camera!). I transferred the recording to CD for him, and commented that the MD compression artefacts were fairly obvious at certain moments (as it happens, also during sibilants, but this time in singing not speech. There are good signal processing reasons why sibilants of relatively long duration are difficult for ATRAC or MP3). My friend didn't know what I was talking about until I taught him what to listen to, and then he, too, found them obvious and could reliably pick them out. I shouldn't have done this really - it turned a recording he was very pleased with into one he was a bit miffed with. I'm afraid I decided not to buy an MD recorder at all on the basis of hearing this recording. Regards, Christopher Hicks ----------------------------------------------------------------- To stop getting this list send a message containing just the word "unsubscribe" to [EMAIL PROTECTED]
