Hi all,

I need to encode 5 wavs (48kHz,16bit,mono) to average bitrates 32kbps and 48 kbps in AAC format. Any advice for switches,profiles etc.? I have Psytel v2.15.

If possible, can someone provide switches for both the resampled (+fs,+cutoff) and nonresampled case?

BTW, the wavs are 'standard' testfragments such as castanets, celine dion, harpsichord, female english speech and, of course, eddie rabbit

Thanx in advance,

Petracci

What exactly do you mean by "average bitrates"? Is it OK if they differ 10-15% from your 32 and 48 kbps or not? This is important, because Ivan Dimkovic just stated in the other thread about AAC+ that there are some bugs with constant bitrates at low bitrates in PsyTEL v2.15. So it would be better if you choose -tape, -radio or -qvbr <x> as the main presets on the command line rather than -br <x>. Furthermore the resulting bitrate differs a lot depending on the material to encode, as you might already know. Perhaps it's necessary that you do some test encodings for your samples, before you decide which VBR setting might fit your needs the most.



If you use any VBR setting or preset, the cut-off is fixed in PsyTEL, so there's no use trying to define it with -c or -fb. Resampling at these low bitrates is always better, at least to my ears - see the thread mentioned above. But 48 kbps for a mono signal is already quite high, so maybe you could try it wíthout resampling. My best command line trying to compare it to FhG AAC, WMA9 and mp3PRO was "-radio -resample 32000", but it gave me 73 kbps for a stereo signal, while the others only needed 64 kbps. You might also try "-qvbr 17 (or lower) -resample 32000 (or 24000)", if this comes out as 32 kbps for your mono signal.

Thanx for your reply..


A margin of 5% is OK here...strange thing is that CBR works fine for me, margins of 1-5%, but trying something like -vr -br 32 gives me over 54 kbps. -vr -b 16 just crashes.

Yeah I now, but I'm comparing with an experimental codec that does not resample. I'm also interested to see what artifacts will popup if resampling is not used at these bit rates...


Where did you get the -qvbr switch from? I did not see that in the Psytel manual...

That's right, this is another bug I've already mentioned in other threads, too. This concept of a "base" bitrate together with -vr or -vbrhi also needs some fixing, if you use it to draw the bitrate down to very low settings. Like I recently wrote to The Belgain, it nevertheless works without problems at high bitrates and is the only way to define an own cut-off frequency with a variable bitrate in PsyTEL.

By the way, the bugs Ivan mentioned in the context of CBR were not critical, so they did not crash AACEnc, but they have an influence on the sound at these low bitrates. That's why it's better to use VBR settings instead.



Then you might also use LAME with --alt-preset 64 --resample 44...



It's mentioned in the index.pdf that comes with the complete package of AACEnc and in the "longhelp" that Ivan once gave here in this board. But I also overlooked it the first time and found it at http://www.soundexpert.info/, where Serge Smirnoff mentioned all switches of every codec he used for his test samples.

By the way, I just found out that you can't resample to 24000 Hz with -qvbr, because the lowest setting would be 0 which gives you a fixed cut-off of 12000 Hz, and this results in another crash because of a "dividing error". So the only usable resample rate is 32 kHz.