My apologies in advance if this topic has been done to death - I searched and found nothing quite addressing the following question:
Does level affect Lame's encoding quality? I understand completely what normalizing does and why it (generally) should be avoided, but here's an example...
When archiving a DVD to XviD, say I want to also backup the commentary track... Often the commentary tracks are extremely quiet, peaking around -10, average being obviously much lower. I don't care about the low level since that's what that big round knob is for, but since I don't want to waste space on a voice-only commentary tracks, I wish to archive at 96kbs - Will LAME's psychoacoustic model behave differently at those low levels, or should I Normalize for a better quality encoding at these low bitrates (96kbs - 128kbs)?
Thanks!
Demtation
In theory this should not matter. But of course there's a possibility of bugs...and I remember seeing threads about this before. Use search and see if you can find them.
IIRC it was more of an issue for VBR, where much earlier lame versions would seriously reduce the bitrate for quieter versions of the same material. The floating ath in later lame versions reduced (but didn't eliminate) the issue.
It's hard to imagine it making any significant difference on a CBR commentary track, but you should always test. I don't recall anyone testing low bitrate CBR in previous threads.
Cheers,
David.
P.S. Here's the thread - it's in the FAQ filed under various - the original question is specifically about replaygaining tracks to a lower volume before encoding, but the whole encoding/volume issue is discussed...
http://www.hydrogenaudio.org/forums/index....showtopic=10637 (http://www.hydrogenaudio.org/forums/index.php?showtopic=10637)
I you apply dynamic compression when playing your track, then yes it would be better to apply this processing before encoding.
This is not a bug but a normal behavior of single pass encodings: if we want to use the ATH, we have to choose its base level. For this choice, we are assuming that the input is in "regular" conditions.