Skip to main content
Topic: objective measurement of HE AAC (Read 3708 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

objective measurement of HE AAC

Hi,
is there any objective measurement of HE AAC?
Until now, I only know the subjective assessment.
thanks

objective measurement of HE AAC

Reply #1
If you're dealing with non-transparent bitrates (which may be the case with AAC at bitrates where HE is used), you would want to do an ABC-HR test to rank the HE AAC against other codecs (or AAC encoders) at similar bitrates to determine which one sounds the best. ABX is only used to determine if the lossy version is transparent when compared against the original.


objective measurement of HE AAC

Reply #3
If you're dealing with non-transparent bitrates (which may be the case with AAC at bitrates where HE is used), you would want to do an ABC-HR test to rank the HE AAC against other codecs (or AAC encoders) at similar bitrates to determine which one sounds the best. ABX is only used to determine if the lossy version is transparent when compared against the original.



I am working with low bit rate HE AAC.
I mean to measure the quality without doing listening test. As far as I know, ABX is also a listening test.
For example, PEAQ, but it doesn't work so well to represent the quality.

objective measurement of HE AAC

Reply #4
Quality is a subjective measure in this case, so it is hard if not impossible to measure without listening tests
Music: sounds arranged such that they construct feelings.

objective measurement of HE AAC

Reply #5
Quality is a subjective measure in this case, so it is hard if not impossible to measure without listening tests

but there are some ways to do objective quality evaluation


objective measurement of HE AAC

Reply #7
I'd say that "objective quality" is akin to measuring the mathematical properties of the audio, but that's not what we're interested in. We're interested in how the audio sounds, and the human ear is the only tool capable of measuring that property of the audio.

The same holds true for video: you can use PSNR to make an objective mathematical assessment of how closely a lossily compressed image corresponds to the original, but PSNR doesn't correspond with which images people tend to think actually look better. Since the thing we care about is which compressed images look the best, PSNR is not the best measure of picture quality.

objective measurement of HE AAC

Reply #8
Quality is a subjective measure in this case, so it is hard if not impossible to measure without listening tests

but there are some ways to do objective quality evaluation


HE AAC is a lossy codec, and one that doesn't really aim to be transparent (the HE is for High Efficiency, meaning it's meant to get "good enough" quality into a very low bitrate stream or small size file). It is built on psychoacoustic principles and listening tests. So if you compare it to a WAV or FLAC file (lossless) and look at their spectrograms, you will of course see that the HE-AAC discards a lot of information, particularly in high frequencies. But this doesn't tell us how it will actually sound, particularly given that HE-AAC is usually reserved for audiobooks or other speech-based recordings that do not use the extremes of the frequency spectrum.
Even with AAC-LC at high bitrates, the spectrogram would again show that some frequencies are being attenuated. But that doesn't tell us whether or not you could distinguish between the two in a listening test such as the popular ABX.

objective measurement of HE AAC

Reply #9
HE AAC is a lossy codec, and one that doesn't really aim to be transparent (the HE is for High Efficiency, meaning it's meant to get "good enough" quality into a very low bitrate stream or small size file).

... or in other words, HE-AAC uses parametric tools which don't preserve the original waveform(s) of the coded signal. Neither does error concealment (since it can only estimate the actual waveform).

And this is what causes problems with objective measures like PEAQ.

Quote
It is built on psychoacoustic principles and listening tests.

So are objective measures like PEAQ (which was trained to reflect a large set of listening test scores). This works sufficiently well at high bit-rates where parametric
coding is not used (I think in Muaddib even improved it once during his Master's thesis, here is the link). But HE-AAC was finished after PEAQ had been standardized,
so it doesn't have any measures for non-waveform-preserving coding (it compares against the original waveform and finds large errors = bad scores, even though
the parametric coding may sound very good).

So, I don't know of any tool for meaningful objective measurement of HE-AAC.

Chris
If I don't reply to your reply, it means I agree with you.

objective measurement of HE AAC

Reply #10
HE AAC is a lossy codec, and one that doesn't really aim to be transparent (the HE is for High Efficiency, meaning it's meant to get "good enough" quality into a very low bitrate stream or small size file).

... or in other words, HE-AAC uses parametric tools which don't preserve the original waveform(s) of the coded signal. Neither does error concealment (since it can only estimate the actual waveform).

And this is what causes problems with objective measures like PEAQ.

Quote
It is built on psychoacoustic principles and listening tests.

So are objective measures like PEAQ (which was trained to reflect a large set of listening test scores). This works sufficiently well at high bit-rates where parametric
coding is not used (I think in Muaddib even improved it once during his Master's thesis, here is the link). But HE-AAC was finished after PEAQ had been standardized,
so it doesn't have any measures for non-waveform-preserving coding (it compares against the original waveform and finds large errors = bad scores, even though
the parametric coding may sound very good).

So, I don't know of any tool for meaningful objective measurement of HE-AAC.

Chris

thanks Chris, so if parametric tools are not used, I mean for mono mode at low bit rate only with AAC+SBR being used, it is also problem for objective measurment because objective measurment mainly compares wavforms.


I was planning to do both objective measurment and subjective listening tests to show the improvement by error concealment.
But now it seems even though the PEAQ ODG values have been improved by error concealment, that doesn't mean the audio quality is improved, right?
In other words, if the ODG value is low, that doesn't mean audio sounds bad, and similarly, if ODG value is high, that doesn't mean the audio quality is good.
so maybe subjective listening tests is the only way to show improvement.

 
SimplePortal 1.0.0 RC1 © 2008-2019