Once we put all of these different tools together to address the data rate, either for transmission, or for retrieval, you end up with many many choices, right? And as we saw in the miracle 13 dB demo and the JPEG images, the signal to noise ratio does not always tell the truth. So we need to have some way of controlling the quality. So for the wide band audio from what we saw before, you can deduce from, you can get certain result in data rates of about 8-12 kbps. And for CD quality audio, you can have anywhere from 64-256 kbps in one of the newer AEC codecs even at 128 or 192, you have gone beyond CD quality what people call real transparent quality. [COUGH] Then for still images, you can get anywhere from 30 to 300, I think should have been kilobytes not megabytes. I have an error here. Based on what quality you use for your JPEG compression. And similarly for about 412 to 1 kbps so five megabits per second you can a very decent video quality for streaming over the internet and the resultant data size depends on the content itself to a large extent. If you have Talking Heads, you can get much better accomplishing ratio, as opposed to Federer and Jokovich playing tennis. Then, there's lots of motions, and sometimes motion breaks. You need to send more ifibs. On the audio side, [COUGH] the quality evaluation tends to be a lot more tricky than the that we used for speech. So now you are having, asking more pertinent questions that relate to the transparency of the signal. And for one part you need real expert listeners who have had many years of training in order to evaluate the codex. And one of the testing paradigms is ABX. You're told A is original, and B is either encoded or B is definitely not B encoded and decoded. And, you're given a third segment. Your task is to see if X is very close to A, or very close to B. So, sometimes it will be the original, sometimes it will be the encoded. And the other one is ABC HR. HR stands for hidden reference. A and B are the original and decoded in random order. And C is always the original. So you have to actually Identify which is the process signal and then assign a subjective scope. In this course, are usually 0.5 scale. [COUGH] And some of the ITU, International Telecommunications Union, they're this tester define. They allow fractional values so that you cover a much bigger scale. MUSHRA is multiple stimuli with hidden reference and anchor. So in this test, you have a lot more content that the subjects are looking at. And you have three or four candidate codex and you throw in codex for which we already know the quality like g.726, g.711, or for very high quality codex some of the g.722, 3's that runs at much higher pitching. And the task of the users, usually you have multiple sets of listeners that listen to this. And it's almost an art implementing and running these tests. So it's a very, very time consuming and elaborate process, but they have to go through all of these when they release the new version of mp3 codec, AAC codec, surround sound, HE-AAC which has the sub-band application. Well, I didn't talk about that in the technology, so let's leave that. So, there is always a need for a subjective tool to kind of quickly assess, especially during the algorithm development, and one of the notable ones for this objective quality measurement is peak. Perceptual evaluation of audio quality. And this branch of research is almost a science by itself because you take the original signal and you take the process signal, and you implement the cyte acoustics just the way some recordings do. And you extract various sets of features and combinations and outcomes, and MOV, this is an acronym for Model Output Variables. Then you put them all together and come up with a grade, and we have used this tool in my work in the past. And, as a first cut, it's very very good. It takes us through the initial experimentation as we are trying to select a specific codec, specific data rate, or a new codec algorithms, and there's some commercial versions, and there's an University version that are free on the internet.
