ATRAC3+ Hi-SP (256kbps) vs. 192kbps

[grenert]

According to Ishiyoshi's data:

http://forums.minidisc.org/index.php?showtopic=11325

There is a high frequency cutoff at around 18kHz with ATRAC3+ 192kbps, while Hi-SP (256kbps) has no cutoff. This makes me wonder if 192kbps might actually provide better quality. Since we hear 18kHz+ poorly if at all, I'm not sure we'd perceive much of a loss from the filter, and then there would be a lot less information to encode. Admittedly, 192kbps is a lot less data as well, but maybe getting rid of the very highest frequencies "cleans up" the material to encode at a lower bitrate with fewer artifacts. I have to admit I don't think I hear artifacts with Hi-SP as it is.

Any opinions on this? Has anyone done much testing?

[Sparky191]

I always thought a lower bit rate will have more artefacts not less.

[greenmachine]

When using the same parameters, like cutoff frequency, yes, but since it is set lower and frequencies above 18 kHz are to a great part inaudible anyway, there is less to encode, the encoder can focus on the important/audible part, a lower bit rate might sound as good or better in theory, the poster's suspicion is justified.

[Sparky191]

I take the point.

Personally i found I prefer the sound of 256 but I wouldn't say there was a huge difference between them, I'd only notice the difference on complex tracks like classical, and then only with very familar tracks and specific sections. My thinking is if you have to go to so much trouble to find a difference, then 99% of the time you won't notice the difference. I don't use my HiMD as a player very much anyway so I don't have a ATRAC library when I do use it, I don't bother with anything other than HiSP or SP.

[syko]

There is just about no difference between 256kbps and 192kbps. Perhaps the only difference you will hear is the stereo accuracy and very slight artifacts in 192k with very high frequencies compared to 256k.

High frequencies require much more data produce accurately than low frequency sounds, so frequency response drops dramatically as bitrates drop. This is because each packet of data is too small and there aren't enough bits for the decoder to replicate the original high frequency sound. This frequency response is not encoder dependent - it's bitrate dependent, so it's not the encoder's fault that it can't produce frequencies that high; it's the file structure that is limiting its ability to reproduce them.

The decoder won't be selective in what it processes. It will always try to process the same sound, and depending on what bitrate is used, will try to stuff as much information into each packet as possible. If it does find a frequency that will not fit into the smaller data packet accurately, most likely it will drop that part (frequency cutoff!). Each encoder will have its own algorithm in figuring out which parts to cut or compress whilst trying to maintain its highest accuracy to the original.

Note also that most mp3s' frequency cut off at 192k is around 16-18kHz, which further justifies that this is not an encoder problem. The higher bit rate will always have the higher quality sound and in no way it will produce more artifacts.

-------------------------------------

Here's something I can compare it to that might help understanding this issue.

"0 1" means one cycle. The definition of Hertz (Hz) is the number of "0 1"s per second - cycles per second.

----------

Here is a data packet representing a sound - 16 bits in its original form.

0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

- This string is fed into the encoder once per second. Assume the sound is at 8Hz because there are 8 "0 1" sequences fed into the encoder per second. This is its theoretical frequency cut-off because there can only be a maximum of 8 "0 1" sequences in one data packet.

----------

Now here's a data packet representing a sound - 10 bits.

0 1 0 1 0 1 0 1 0 1

- Note now that there can only be a maximum of 5 "0 1" sequences. This is the new theoretical frequency cut off of 5Hz for this type of file.

----------

Of course, with something say like 192kbit/s, there are 192000 bits to play with fed into the decoder per second, so things are less crude than my examples. This does not mean the frequency cutoff is 96kHz - sound is never so pure like a square wave as represented by my "0 1" sequence. Music is a combination of 0 and 1s with many waves at once.

This is the very basic idea of an audio file.

[greenmachine]

I think what you're talking about is the sample rate. Since lossless and lossy formats usually use the same sample rate for standard / high quality audio (44.1 kHz), there is no forced cutoff below 22.05 kHz. Although the presets are usually set / tuned differently, you could force the encoder (with LAME, for example) not to filter the highest frequencies even at 192 or 128 kbps. The trade-off would be increased artifacts.