Hi guys,

I've been lurking for a few days now and today I am taking a plunge and asking a question regarding the comparison of AAC encoders and decoders. I did a search to check if there was a thread with the info I needed but I did not find an answer.

From reading on these forums, I have noticed that subjective comparison really seems to be the way to go and that's understandable seeing as this is perceptual encoding using psychoacoustic parameters that vary from one maker's encoder to the other. So I understand that comparing waveforms might not be the best way. However, I wonder if waveform comparison can still give me some type of indication regarding compression/decompression quality. I guess my question is more along the lines of "How bad is it to do waveform comparison?"

Today I performed a short test. I took a very short PCM wav file and encoded it using different AAC encoders. Then I decompressed them using their respective decoders just to see how similar the uncompressed file was to the original template wav file. Am I wrong in assuming that a good encoder-decoder should give a very similar waveform after compression-then-decompression? Or does the psychoacoustic modeling throw a wrench in that train of thought meaning that one of them might generate a final waveform that shows many dissimilarities when compared to the original and still be very close in terms of how it sounds? I do understand that a codec's implementation might remove information that is deemed unimportant without it significantly affecting the perceived audio quality and thus waveform comparison might not be accurate at all.

I am asking because one of the encoders I tested for my special case (very short wav file) did give me a fairly different waveform. It is however very difficult for me to tell the difference perceptually (once again, a very short wav). The thing is I am trying to quantitatively determine if one codec is of acceptable quality when compared to the more popular AAC codecs implementations like iTunes' and Nero's because I don't trust my hearing enough and I don't have enough time to perform all those double blind tests.

Thanks!

When using very high bitrate the error in the encoding should be small so the wave form of the original and the encoding should be close.
When using low to moderate bitrate this is not necessarily so. An encoder can 'throw away' a lot for good reason so that you won't hear the difference. So the perceptual error can be very small though the wave form error can be pretty large.

wav forms are not good for analyzing the quality of an encoder. There really is no way to quantitatively measure how acceptable one encoder is other than by conducting blind listening tests. You can look at wav forms all you want but it doesn't do any good really. In other words, there isn't a visual way to measure audio quality.

Most modern-day encoders should have no issues making similar looking wav forms anyway even at bitrates as low as 128kbps-160kbps. Once you get up to 192kbps then they all sort of look alike.

Might I ask what you are trying to accomplish? Are you trying to make your own AAC encoder or what? You said that you don't trust your ears enough but if you are just trying to decide which encoder to use then the only thing you can trust is your ears. Who else (that you care about) is going to be listening to the content? If you will be the only one that you really care about then go with your ears.

Hi guys, thanks for the replies.

I am having trouble imagining that people at Ahead Software and Apple evaluate the quality of their codecs by performing listening tests. It somehow does not seem professional

I keep thinking that there must be some software/hardware that can perform a fairly accurate analysis of signal degradation and determine the data lost by doing analysis in the frequency and temporal domains. Maybe there are some advanced mathematical models of the human hearing system that could be used in these tests. They would then have a bunch of audio samples being batch tested by their hardware/software solution that would be able to determine the difference with the original template. But then again, I might be totally wrong.

I would be interested in finding out what those means of testing could be so that I can determine what the variations between different codecs are and then determine an acceptable range of variation beyond which a codec would not pass the "quality" test.

Thanks again.

http://www.hydrogenaudio.org/forums/index....showtopic=59835

The whole idea behind lossy audio compression is that by exchanging some of the signal for noise, in a way that is least perceptible to human hearing, you need less information to store it. The ONLY way to assess how well the encoder does that is by evaluating it against...human hearing. Nothing else. No mathematical model will tell you how to trade off, say, lowering your lowpass filter vs more pronounced smearing artifacts.

Except mathematical model that models human hearing good enough