VBR in Nero overshoots or undershoots target bitrate

[MrCommunistGen]

I'm using v1.1.34.2 of the Nero AAC codec in MeGUI. I'm trying to encode a 3hr ac3 audio file and I'm shooting for ~128kbps but I can't seem to get it close. When I use a quantizer of 0.48 I get 112kbps and when I use 0.49 I get 153kbps. I tried modifying MeGUI's job file so that it would use q=0.485 and q=0.489 but it just reverts to q=0.48.
Other than using ABR (which is what I settled for) does anyone have any suggestions. Also, am I right in assuming that an ABR 128kbps will probably retain more quality than a 112kbps VBR?

-mcg

[nurbs]

VBR is not supposed to reach a target bitrate, but a target quality. So the filesize will of course vary depending on the complexity of the content. That said if in your example with 0.48 and 0.49 you encoded the same file the difference in size is a bit strange.

[skromnibog]

1. For different ranges of q values different tables with values for psychoacoustics parameters are used. When you change q from 0.48 to 0.49 the table being used changes and then big difference in bitrate can happen. However what you report is very big change. Can you please upload somewhere short samples (up to 30s) to demonstrate this?
2. VBR mode in Nero AAC will produce target bitrate on big set of different samples. On a single sample produced bitrate may vary a lot. It depends on a specific sample whether ABR 128kbps would sound better than VBR 112kbps. It may happen for example that for some sample ABR produces 128kbps and VBR produces 64kbps, but yet without any difference that can be heard. Then in another example ABR can produce 128kbps and VBR 160kbps and difference could be easily heard. In average, but only in average on many samples, quality of ABR 128kbps can be expected to be better than VBR 112kbps.