FAAC LC-5.1 Encoding Information

[hans-jürgen]

Interesting thread, thanks for coming up with that! I would like to ask if anyone could add FAAC to the list, i.e. encode the same AC-3 "original" at e.g. -q 75 -c 13000 in foobar2000 (with foo_faac, TNS and M/S enabled, output MP4, LC, direct transcoding through piping, no Vorbis in between please). Of course this would also be possible with the 3ivx DS filter suite (free demo for one month, then US$ 19.99 for the audio encoder based on FAAC).

Anyhow, this should result in approximately 225 kbps/6ch VBR according to previous multichannel tests, but could differ of course due to the input sound of that sample. And mentioning the file sizes for all test samples might be useful, too, e.g. to check if all settings really resulted in 5.1 files, also to decide which ones are small enough for a modem download. ;-)

Including 5.1 AAC Low Complexity files in any multichannel transcoding test makes sense, because the first mass-production 5.1 hardware decoders (e.g. in DVD players) will very likely be able to decode only this profile, not HE AAC.

Title Edit: KpeX

[hans-jürgen]

Originally posted by hans-jürgen
Anyhow, these bitrates are also possible with the Low Complexity Profile of AAC (a stereo bitrate of 90-100 kbps VBR is considered to be the "sweet spot" of AAC LC), that's why I asked if someone could test FAAC at a VBR quality of 75% and a cutoff at 13 kHz with your original sample. OK, using foo_faac in foobar2000 with the settings already mentioned (and no DRC in the foo_ac3 plugin) results in a 230 kbps/6ch VBR MP4 file with 48 kHz sample rate. Since I don't have a 5.1 sound card and surround speakers, I had to use foobar's downmixer DSP for playback in order to hear it with my Beyer DT 770 headphones, and to me it sounded very impressing, no noticable flaws without comparing it to the original or other samples. I probably couldn't listen to the WMA9 Pro sample anyhow, because I don't have WMP9 or a DS filter either, and my system is also too slow for the AC-3 and the HE AAC sample.

Doing the downmix while transcoding ("Use DSP" in the disk writer settings) results in a 85 kbps/stereo VBR MP4 file which might be interesting when you really want to save space on a CD or DVD. As the sample is quite "busy" during the 3 minutes (is this a Lord of the Rings trailer?), I guess a whole movie with more quiet parts (dialogues etc.) will even have a lower overall bitrate with these FAAC settings.

[KpeX]

Originally posted by hans-jürgen
OK, using foo_faac in foobar2000 with the settings already mentioned (and no DRC in the foo_ac3 plugin) results in a 230 kbps/6ch VBR MP4 file with 48 kHz sample rate. Since I don't have a 5.1 sound card and surround speakers, I had to use foobar's downmixer DSP for playback in order to hear it with my Beyer DT 770 headphones, and to me it sounded very impressing, no noticable flaws without comparing it to the original or other samples. I probably couldn't listen to the WMA9 Pro sample anyhow, because I don't have WMP9 or a DS filter either, and my system is also too slow for the AC-3 and the HE AAC sample.

Doing the downmix while transcoding ("Use DSP" in the disk writer settings) results in a 85 kbps/stereo VBR MP4 file which might be interesting when you really want to save space on a CD or DVD. As the sample is quite "busy" during the 3 minutes (is this a Lord of the Rings trailer?), I guess a whole movie with more quiet parts (dialogues etc.) will even have a lower overall bitrate with these FAAC settings. Thanks for these results. I've often considered doing some testing to determine the ideal -q setting for FAAC 5.1 channel LC AAC with an optimal combination of quality and small filesize. Also, we should consider if the cutoff (-c) setting needs tuning, or if the FAAC defaults are acceptable for movie encoding. I almost think due to the high speech content of movie soundtracks we could bump the cutoff up to 16-19 khz ( at -q 75-85 ).

[hans-jürgen]

Originally posted by KpeX
Also, we should consider if the cutoff (-c) setting needs tuning, or if the FAAC defaults are acceptable for movie encoding. I almost think due to the high speech content of movie soundtracks we could bump the cutoff up to 16-19 khz ( at -q 75-85 ). The new v1.24 uses automatic cutoff changes with different -q settings (e.g. -q 50 with 10 kHz, -q 150 with 22 kHz), so its usage has become a little bit easier if you don't like command lines. ;) And -q 75 now uses 13 kHz, that's why I set foo_faac to that cutoff (not updated to v1.24 yet) and enabled TNS, too (another change from v1.23.5). By the way, the cutoffs can still be changed individually of course.

If this is a good idea should be tested thoroughly by anyone who plans to do that, because there are several issues related to such a change:

* The overall bitrate will rise (no, not a bug, a feature), because FAAC works this way to ensure a constant quantizer quality with a higher cutoff.

* Speech content hardly ever gets hurt by slightly lower cutoffs than the usual HiFi crowd needs, e.g. look at the cutoffs and sample rates used in speech codecs. And yes, I know what sibilants are... ;)

* There are no public formal listening tests that provide any proof to these kind of "lowpass experts" myths (I don't mean you, KpeX).

* A realistic multichannel movie listening test would have to be done on a surround speaker setup (not on stereo PC boxes or with headphones), because these speakers normally are not "state of the art" in high-frequency reproduction (except for some very rare setups that probably no one here can afford). And it should probably include viewing the picture if possible, because this will draw the attention away from audio artifacts like an assumed high-frequency loss.

* Last but not least, the AC-3 format normally uses channel coupling for frequencies above 15 kHz, i.e. they come in mono, so too much effort to recreate surround images up to e.g. 22 kHz is a little bit exaggerated in my opinion. :)


Going the opposite way in order to lower the overall bitrate even more might be possible, too, but then you should also use the internal resampler DSP of foobar2000 and set it to 24 kHz (half of the original sample rate) which will automatically lower the cutoff to 12 kHz in FAAC. Using -q 75 still sounds acceptable with this sample in my opinion and reduces the bitrate to 69 kbps/stereo or 185 kbps/6ch. But this setting surely is the lowest limit for FAAC (or AAC LC in general), and going higher with -q again will cure the dropouts etc., e.g. the default -q 100 results in 79 kbps/stereo or ~213 kbps/6ch then (with resampler DSP set to 24 kHz of course), so it still comes out considerably smaller than -q 75 and no downsampling.

[KpeX]

@Hans-Jurgen

Thanks for this information :). I'll split these posts now as they are not related to the WMA topic, and this is important information for FAAC users.

[SeeMoreDigital]

I wondered where these posts had gone!

Hans, I guess you used Sagittaires 6Ch AC3 file as a source!

Are you able to post your encode. If not, what file size did you get it to?

Cheers

[hans-jürgen]

Originally posted by KpeX
I'll split these posts now as they are not related to the WMA topic, and this is important information for FAAC users. Good idea, thanks.

[hans-jürgen]

Originally posted by SeeMoreDigital
Hans, I guess you used Sagittaires 6Ch AC3 file as a source! Yes, English.ac3 seems to be the original from the DVD with 448 kbps/6ch and not transcoded with Ogg Vorbis in any way.

Are you able to post your encode. If not, what file size did you get it to? Sorry, I'm a modem user, so this would take too long, and my available web space would be completely gone.

Original AC-3 file ("Lord of the Rings II" trailer, 2:58 min.):
* English.ac3 9,736 KB = 448 kbps

Direct 6-channel transcode with foo_faac and recommended settings:
* English-6ch-r48-q75-c13.mp4 5,059 KB = 230 kbps

Stereo downmix with the same settings:
* English-2ch-r48-q75-c13.mp4 1,892 KB = 85 kbps

Stereo downmix and SSRC resampler DSP at 24 kHz and -q 100:
* English-2ch-r24-q100-c12.mp4 1,753 KB = 79 kbps (~213 kbps/6ch)

Stereo downmix and SSRC resampler DSP at 24 kHz and -q 90:
* English-2ch-r24-q90-c12.mp4 1,668 KB = 75 kbps (~203 kbps/6ch)

Stereo downmix and SSRC resampler DSP at 24 kHz and -q 75:
* English-2ch-r24-q75-c12.mp4 1,527 KB = 69 kbps (185 kbps/6ch)

To estimate the resulting bitrates for the missing 6-channel transcodes I used a multiplier of 2.7 derived from the first two tests (confirmed with the lowest one in the meantime). The bitrates are the values shown in foobar2000's file properties (the sample rate for the resampled files is listed as 48 kHz there due to the MPEG-4 implementation in FAAD2, but in reality is 24 kHz). The file sizes are the ones shown in Windows Explorer.

By the way, if you need to define an exact file size/bitrate for CD/DVD burning and therefore don't want to use the default VBR mode of FAAC, you can also use ABR with -b which will stick to the desired bitrate quite closely, but doesn't sound as good and wastes bits of course on quiet passages.

There's also a guide with screenshots on how to use foo_faac for AC-3 transcoding (enabling resampler and downmixer DSP etc.) over at Everwicked.com in their online XviD/DivX Guide, Chapter "Audio transcoding". It is a little bit outdated, because it used foobar v0.7x, so the current GUI looks slightly different, but the important facts are still correct - except using the Main profile which should have been replaced with the LC profile of course:

http://www.everwicked.com/content/XviD_Guide/content-07.php

Last but not least the new Winamp output plugin based on FAAC v1.24 has been compiled and uploaded to RareWares.org by John33 yesterday, so it should be possible to do direct AC-3 -> AAC/MP4 transcodings with Winamp now, because there is also an AC-3 input plugin. I don't know if you could also resample and downmix while transcoding though.

The same is true for CoolEdit/Adobe Audition, because the new cool_faac.flt has been reported to work OK, too, if you make sure that you set the options manually (the automatic configuration seems to use wrong defaults).

[KpeX]

I did some FAAC encoding tests with 1.24 beta and a 9-minute 5.1 AC3 sample from Revolutions. I first decoded the ac3 to six channel wav with Besweet/azid (using light DRC and normalization) and encoded the resultant six channel wav with a variety of settings:


Quality Bitrate (kb/s) Bandwidth (hz)
55 185 10600
60 194 11200
65 205 11800
70 216 12400
75 238 13000
80 249 13600
85 258 14200
90 268 14800
95 293 15400
100 305 16000
105 317 16600
115 336 17800
120 362 18400
125 374 19000
130 385 19600From these results, I can conclude that a range of q between 75 and 100 is usually acceptable for good quality and medium filesize. Later, I may do some testing with tweaking -c (bandwidth) settings.

[hans-jürgen]

Originally posted by KpeX
I did some FAAC encoding tests with 1.24 beta and a 9-minute 5.1 AC3 sample from Revolutions. I first decoded the ac3 to six channel wav with Besweet/azid (using light DRC and normalization) and encoded the resultant six channel wav with a variety of settings:
[...]
From these results, I can conclude that a range of q between 75 and 100 is usually acceptable for good quality and medium filesize. Later, I may do some testing with tweaking -c (bandwidth) settings. Thanks for the additional testing, did you also listen to the files? By the way, using DRC will usually increase the overall bitrate of a VBR codec a bit, because the quiet passages cannot be encoded as efficiently as without compression.

[KpeX]

Originally posted by hans-jürgen
Thanks for the additional testing, did you also listen to the files? By the way, using DRC will usually increase the overall bitrate of a VBR codec a bit, because the quiet passages cannot be encoded as efficiently as without compression. I haven't done much listening to the files yet, although in a quick initial comparison I didn't notice much difference from the original on my 5.1 system until the quality was below 75. I hope to test more soon comparing DRC settings and tweaking -c settings.

[hans-jürgen]

Originally posted by KpeX
I haven't done much listening to the files yet, although in a quick initial comparison I didn't notice much difference from the original on my 5.1 system until the quality was below 75. If you go below -q 75 (combined with 13 kHz cutoff), you should not forget to downsample the input file to 24 kHz, too, because this increases coding efficiency, so you can raise -q again and still get a lower overall bitrate (see examples above). BeSweet uses the same high-quality SSRC resampler as foobar2000 for that, as far as I know. And probably WMA9 Pro also uses downsampling with low bitrate settings (can't check this though).

I hope to test more soon comparing DRC settings and tweaking -c settings. I just did another test with DRC enabled in foo_ac3.dll, but the 6-channel MP4 file came out with the same bitrate (230 kbps) as without DRC. This may have several reasons, e.g. that there is no Dynamic Range Compression used by Dolby/the sound engineer on this AC-3 track, or simply too few chances to make a difference. Furthermore I don't know the DRC level that is used in foo_ac3 (there are no options, only on/off).

By the way, there is an update available for foo_dsp_extra.dll that enables downmixing to Dolby Surround or ProLogic II in addition to the normal 5.1 to stereo downmix. You can find the link in the foobar forum on HA.