384kbps ac3 vs 640kbps ac3

[robby51]

I have two mkv files; one with 2500kbps video and 640kbps audio, and another with 2150kbps video and 384kbps audio.


The mkv with the 2150kbps video looks consistently better (the 2500kbps has slightly more detail in dark areas, but has slight distortion of features in light areas, as well as quite a bit more noise), so I guess the 2150kbps video had a more skilled encoder.


I was considering taking the 2150kbps video and muxing it with the 640kbps audio, but I have realized i'm not sure how to tell which audio file is better (I don't know if you can mess up an ac3 encode like you can a x264 encode). The original source of both audio files was a 1500kbps dts.


So I got an audio spectrum of both audio files and I realized I have no idea how to read it. The 384 cuts off at 14.5khz, but the 640 has horizontal lines for minutes at a time; is missing some thin green lines that the 384 has, and has a bunch of stuff above 20khz that looks out of place. Is there anything that suggests the 640 may be of lower quality by looking at the diagrams?


384: http://i1277.photobucket.com/albums/y489/test124124/384_zps901715c2.jpg
640: http://i1277.photobucket.com/albums/y489/test124124/640_zps43faef09.jpg

[turk690]

The better audio is the one whose sound you like more, no?
That said, these are spectral frequency displays. Representing audio graphically like this is helpful in having an idea how much energy is clustered in what frequency groups. For example, if I were capturing audio from an LP record, this display is preferable because clicks are bright solid vertical lines (maximum energy over the whole frequency spectrum, therefore easy to identify and remove).
But for what you want, I'd prefer a waveform display (vertical axis is signal level, center line is ∞). A glance at this will give you an idea of average levels and peaks; coupled with a frequency analysis display that will show frequency response at any instant, time-wise.
The 384 cutting off above about 15kHz is typical of files downloaded off the internet; site audio encoders low-pass like this to prevent unpredictable aliasing artifacts the closer content goes to 20kHz. The 640 looks more like, say, what I would rip straight off an audio CD; there is energy up to 20kHz (or more, as theoretically it should be 22.05kHz without lowpass, or even 24kHz if it came from DVD).

[hogger129]

Provided you are encoding both from the lossless source, 640kbps should sound the best. The one with 2150kbps is likely the lossless track. As for the frequency cutoff for the 384kbps track, can you even hear that high? I probably can but I'm 29 years old. 384kbps isn't terrible, but the higher rate should sound better. I usually use 448kbps as a happy medium. It uses less space than 640kbps does, but it sounds fuller and more detailed than 384.

[Mephesto]

That's the worst spectrograph I've ever seen. Give us a short sample of both ac3s. I'll take a look with a high quality spectrograph and see if any have DCT degradation on the lower shelves which is a lot more important than a cut-off at the upper frequencies that you likely won't hear if you're older than 15 anyway.

Is this stereo or surround-sound btw? AC3 is really bad and I wouldn't trust bitrates below 256 for stereo so if your audio is surround-sound... *shudders*