Hi.
I just tried to encode-decode the same sound file 10 times with the same quality settings and it gave me interesting result. All the artefacts became very loud and obvious. Each encoder has shown it's own unique kind of losses. And now I wonder if there is such software which encode-decode a sound file cyclically and shows frequency spectrums and probably more information about losses accrual. Or perhaps there are some devs who's interested in this idea?
Full Version: Transcoding looping
Welcome Bubba! 
Interesting first post... and brings to light a technique I have used prior to now when performing listening tests. If I cant hear any artifacts, I'll re-encode a few times, until I know where problems are likely to occur, and then focus on that section. This idea formed from (Den? / 2bdecided?) who drop the quality of the encoding right down to begin with, in order to find out where problems are likely to occur.
Good to bring this to the table, hope a dev will pick it up
Interesting first post... and brings to light a technique I have used prior to now when performing listening tests. If I cant hear any artifacts, I'll re-encode a few times, until I know where problems are likely to occur, and then focus on that section. This idea formed from (Den? / 2bdecided?) who drop the quality of the encoding right down to begin with, in order to find out where problems are likely to occur.
Good to bring this to the table, hope a dev will pick it up
If you re-encode a lossy file over and over again, how will you know (given the artifacts produced) that it will spawn newer artifacts based on previously generated ones? If this is the case, it would not be a good way to get an idea of where artifacts are likely to occur for certain codecs, or maybe it is? Comments?
QUOTE(ChangFest @ Apr 22 2004, 06:48 AM)
If you re-encode a lossy file over and over again, how will you know (given the artifacts produced) that it will spawn newer artifacts based on previously generated ones? If this is the case, it would not be a good way to get an idea of where artifacts are likely to occur for certain codecs, or maybe it is? Comments?
Hmm... Well, if we got all transcoding step files we can do an analysis of them. Also we can try to do delta analysis as well. Anyway after many transcoding cycles the sound becomes very distorted, even with high bitrate. You even can use your ears to do the analysis, but with table/graph it would be more detailed.
The problem with using graphs for audio encoding analysis is that it's not very reflective of the actual sound quality. An encoded audio track may provide a graph that looks very close to the original (unencoded) track, but have comparatively poor sound quality. While another encoded track may provide perceptualy transparent sound quality while generating a graph that looks very different than the original track.
A graph may provide some discreet info about how specific frequencies are treated with re-encoding, but not the overall resulting sound quality. For that, a comparative listening test is the best tool, in my opinion.
I wonder if this kind of repeated re-encoding may generate anchors for codec listening tests with an additional benefit of artifact training? If we use a codec being tested to re-encode an anchor, say, ten times in series, it may help testers better identify artifacts in other encodings, at least in the same format the anchor was encoded in. And if not, then at least it should provide a usable basic anchor for the test.
A graph may provide some discreet info about how specific frequencies are treated with re-encoding, but not the overall resulting sound quality. For that, a comparative listening test is the best tool, in my opinion.
I wonder if this kind of repeated re-encoding may generate anchors for codec listening tests with an additional benefit of artifact training? If we use a codec being tested to re-encode an anchor, say, ten times in series, it may help testers better identify artifacts in other encodings, at least in the same format the anchor was encoded in. And if not, then at least it should provide a usable basic anchor for the test.
QUOTE(ScorLibran @ Apr 22 2004, 02:15 PM)
The problem with using graphs for audio encoding analysis is that it's not very reflective of the actual sound quality. An encoded audio track may provide a graph that looks very close to the original (unencoded) track, but have comparatively poor sound quality. While another encoded track may provide perceptualy transparent sound quality while generating a graph that looks very different than the original track.
A graph may provide some discreet info about how specific frequencies are treated with re-encoding, but not the overall resulting sound quality. For that, a comparative listening test is the best tool, in my opinion.
A graph may provide some discreet info about how specific frequencies are treated with re-encoding, but not the overall resulting sound quality. For that, a comparative listening test is the best tool, in my opinion.
Actually, i don't think we need to know about the actual output quality, mainly because after 10 times of re-encoding it's gonna be total crap with any encoder. But with graphs (and again, probably more characteristics) we can see the distortion tendencies and overall the speed of the "crapping" of each of encoders..
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.