Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: FAAC vs LAME MP3 at about 128 kbps? (Read 16310 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

FAAC vs LAME MP3 at about 128 kbps?

Hello everyone,

I recently bought a Nokia 5800 Xpressmusic phone that came with a 8 GB microSD card. My main portable music player is a Rockboxed Sansa e280 filled with q4 Oggs and I'm extremely happy with it - I'm probably not going to use the new phone much as a music player, but occasionally anyway.

The 5800 music player doesn't support Ogg Vorbis so I'm left with AAC or MP3 as a music format - I'm not willing to sacrifice much more than 3-4 megabytes per song, so 120-140 kbps bitrates would be nice (and I've been very happy with how my Vorbis files sound at similar bitrates). Is there a consensus on whether FAAC-generated AACs are better than LAME-generated MP3s at those bitrates, and if so, with what kind of settings (and encoder versions, if there has been recent development)?

I'm running Linux so using other AAC encoders than FAAC is probably not an option (correct me if I'm wrong). Also, if there would be any other difference between using AAC or MP3, such as battery life, I'm interested. Thanks.

FAAC vs LAME MP3 at about 128 kbps?

Reply #3 –
Quote from: crwl on
Is there a consensus on whether FAAC-generated AACs are better than LAME-generated MP3s at those bitrates, and if so, with what kind of settings (and encoder versions, if there has been recent development)?


Not that I know of, but listening tests like some of these may help you. Linked from that page, the multiformat test in 2004 shows that the chosen settings for LAME v3.96 ~128kbps and iTunes AAC 128kbps were statistically tied, both being beaten by (AoTuV) Vorbis and Musepack settings chosen for similar bitrates.

In the preceding test, AAC test (v2), also in 2004, Nero was possibly better than FAAC (though with less than 95% confidence), and probably not as good as iTunes AAC (again <95% statistical confidence).

Perhaps that implies that FAAC was probably inferior to LAME at the time.

Since those tests, I understand that of the two AAC encoders you can consider for Linux, Nero AAC encoder has had a good deal more tuning, especially on VBR modes, so it's likely to be closer to iTunes than in 2004 and I'd guess it might be clearly better than FAAC.

Equally, LAME VBR has been well tuned, and has recently been statistically tied, perhaps surprisingly, with Helix MP3 at first place in a 128kbps MP3 test.

Both LAME 3.98 and Nero AACenc in VBR mode should be excellent candidates for portable listening at around 120-140 kbps. mp3gain and aacgain can also be used to adjust the playback volume to Replay Gain level. The vast majority of artifacts will be not-annoying (4.0 on the impairment scale, 5.0 being transparent)

If there's nothing much in it, I'd be tempted to choose MP3 just because it's almost universally supported and you can easily transfer it to other devices. Lame -V5 should be in the bitrate range you specified on average.
Dynamic – the artist formerly known as DickD

FAAC vs LAME MP3 at about 128 kbps?

Reply #4 –
Slightly OT, but it sounds like the process you are about to undertake will be a perfect time to consider this:

IF you find MP3 is suitable for you at those bitrates you will gain battery life on your Rockbox'd Sansa by switching it as well to MP3 (assuming a recent build).
Vorbis and MP3 decoding were always close in terms of efficiency on your Sansa but then saratoga made MP3 use both cores and power consumption has dropped significantly.
Creature of habit.

FAAC vs LAME MP3 at about 128 kbps?

Reply #5 –
You also might want to think about HE-AAC at bitrates around 48kbps.  Many cellphones today can properly playback HE-AAC files and that would allow you to get quite a bit of storage out of your 8GB card.  I just don't know if RockBox properly supports HE-AAC or not.  Still, you could get over 4,000 tracks onto your 8GB card when using 48kbps HE-AAC (this is with the assumption that each track is 4 minutes in length).

FAAC vs LAME MP3 at about 128 kbps?

Reply #7 –
Thanks for your suggestions. If FAAC at ~128 kbps won't be better than LAME at -V 5, then I'll probably stay with MP3 on the phone. The current transcoding solution I'm using, Amarok 1.4 with the transKode script, doesn't seem to support any other AAC encoder than FAAC, so I'm probably not currently willing to change to the Nero encoder at the moment.

Hm, I could maybe create a wrapper script that pretends to be a FAAC executable but actually executes NeroAacEnc and NeroAacTag... that shouldn't be too difficult.

Quote from: Soap on
Slightly OT, but it sounds like the process you are about to undertake will be a perfect time to consider this:

IF you find MP3 is suitable for you at those bitrates you will gain battery life on your Rockbox'd Sansa by switching it as well to MP3 (assuming a recent build).
Vorbis and MP3 decoding were always close in terms of efficiency on your Sansa but then saratoga made MP3 use both cores and power consumption has dropped significantly.


I noticed this but am not probably going to change my choice of format from Vorbis to MP3 on that device (unless I take the time to do some ABX tests and see if I really notice any difference between them at similar bitrates)... I'm having >15 hours of runtime even with Vorbis files, which is plenty

FAAC vs LAME MP3 at about 128 kbps?

Reply #8 –
Found this thread through Google, after searching for answers on my question. There seems to have been ~1½ years without any posts, but the stuck thread is understandable since FAAC is stuck at v1.28. Anyways, after reading this thread it seems much based on quality comparison on music. Anyone got news on more like voice stuff? Eg. I tape lots of stuff with DV. I'm usually encoding with LAME (either --cbr 112 or --abr 128). Any bitrate beyond this seems superfluous to my needs. Anyways, I'm thinking about changing to FAAC. Would I notice any difference with faac -b 128? Worse or better? Anyone with real-life experience?

Ty in adv~
Regards~

FAAC vs LAME MP3 at about 128 kbps?

Reply #9 –
Quote from: Gew on
Found this thread through Google, after searching for answers on my question. There seems to have been ~1½ years without any posts, but the stuck thread is understandable since FAAC is stuck at v1.28. Anyways, after reading this thread it seems much based on quality comparison on music. Anyone got news on more like voice stuff? Eg. I tape lots of stuff with DV. I'm usually encoding with LAME (either --cbr 112 or --abr 128). Any bitrate beyond this seems superfluous to my needs. Anyways, I'm thinking about changing to FAAC. Would I notice any difference with faac -b 128? Worse or better? Anyone with real-life experience?

Ty in adv~
Regards~


FAAC is not really a complete AAC encoder.  Its more of an example of how an AAC encoder could work then something you would actually use to encode music.  I would expect it to produce substantially worse results then any normal MP3, AAC, Vorbis, or WMA encoder.  But if you're still unsure after reading the above posts, you should probably try it yourself.

FAAC vs LAME MP3 at about 128 kbps?

Reply #10 –
I was in a similar situation a few months ago because I wanted to rip and encode some audiobook CD sets for listening on my Rockboxed Sansa Fuze/iRiver H340/iRiver H140 and perhaps also on my Nokia N810.  Initially I tried out lame, ogg vorbis, speex and faac, ideally aiming for bitrates 64k or lower and encoding mono at 22050.  I quickly found that speex really is designed for transmission of speech and that's where its advantages lie - it can't compete with regular lossy codecs outside of that arena as far as I can tell; it sounded terrible next to all the others, no abx necessary.  Next to go was faac; truly terrible at very low bitrates and no abx required. This prompted me to try neroAacEnc and it was great at low bitrates.  Lame and Ogg Vorbis were also very good, so here were three encoders which each produced nice results at very low bitrate when encoding speech.  I settled on lame at 48k, 22050, mono (lame options --resample 22.05 -m m --abr 48).  It's way better than I'd have guessed and produces perfectly acceptable and natural sounding reproduction of the human voice for this purpose, though I definitely wouldn't use it for music at these settings.  Lame won out for me over Nero and Ogg Vorbis for two reasons, first both the command line rippers I use (abcde and ripit) allow me to set the encode options easily (same with Vorbis but not possible to integrate Nero into these rippers  at present) and secondly all my devices will play it (same with Nero's aac, but not with Vorbis...kind of...the Nokia tablet can play Ogg but decodes in software not via the DSP so it's best avoided).

Anyway I'd suggest avoiding faac for low bitrates and go for Ogg, Lame or Nero.

FAAC vs LAME MP3 at about 128 kbps?

Reply #11 –
Quote from: Takla on
Lame won out for me over Nero and Ogg Vorbis for two reasons, first both the command line rippers I use (abcde and ripit) allow me to set the encode options easily (same with Vorbis but not possible to integrate Nero into these rippers  at present)


I don't think "i use some crappy software for ripping" is a good argument against Nero.  Theres tons of good stuff out there that will work with command line encoders.  EAC, foobar, CDex, dbpoweramp, etc. 

Quote from: Takla on
the Nokia tablet can play Ogg but decodes in software not via the DSP so it's best avoided).


Pretty much everything on earth decodes MP3 and Vorbis via CPU.  DSP decoders offer no real advantage over CPUs for audio.  I doubt [random device] is any different then everything else.

FAAC vs LAME MP3 at about 128 kbps?

Reply #12 –
Quote from: saratoga on
I don't think "i use some crappy software for ripping" is a good argument against Nero.  Theres tons of good stuff out there that will work with command line encoders.  EAC, foobar, CDex, dbpoweramp, etc.



Wow.    That's a rather more aggressive tone than I'd ever expect in reply to such an innocuous post.  I'm sorry if my post irritated you but what to do?

I made no argument against Nero.  I thought Nero was excellent,  but so was Lame and so was Vorbis and in the end Lame is simply more convenient for me for this particular task.  YMMV.  All were more than acceptable in terms of quality at the low bitrates I wanted.  So I made my choice for a couple of reasons which aren't about audio quality or integration with any of the apps you mentioned, and didn't make or imply any criticism of Vorbis or Nero, both of which I use for different tasks.  I use Vorbis for almost all my music collection for my PC, and Nero (with mplayer and named pipe) for transcoding dts or 6 channel ac3 to aac for my DVD rip/encodes.  I'm happy with both, happy with Lame and happy with the rippers I use.  Lame fits best all around, so I chose it.  YMMV.  The only encoder I thought really poor for low bitrates was faac (speex as well, but that was my mistake for using it for something other than its intended purpose, so not of a criticism of it, but of me).  If I was really keen to use Nero with Ripit I could; it's a Perl script so not exactly rocket science to modify to have it call ~/bin/neroAacEnc instead of /usr/bin/faac....but I'm satisfied with Lame so I didn't.

I think you got my mention of command line stuff somewhat back to front.  The reason Nero isn't in the mix for my speech encodes is not that I have some trouble figuring out how to pass Nero options to a Windows GUI app, as I'm not using Windows.  I use a couple of command line rippers, usually on a headless machine (set up so I push a CD into the slotloader and it rips and encodes, or alternatively via ssh if I want to do something other than the usual automated rip and encode to flac).  Neither ripper is crappy, both ripit and abcde produce bit identical files as EAC with AccurateRip when I tested them (rip+encode to flac, decode to wav, make sha1 or md5sum, compare) and can use many different encoders but right now Nero isn't one of the supported encoders in the official builds and I doubt it will become one.  Free(dom) software projects certainly support a wide range of codecs but generally avoid depending on proprietary encoders, so at the moment aac encoding is usually done via faac (or ffmpeg).  I know someone made a patch to get Nero supported in abcde but I didn't pay a lot of attention.  Like I mentioned before Ripit is a perl script so simple to modify to use Nero if required.  But Lame is doing exactly what I wanted, extremely well, works fine, I'm satisfied.

Quote from: saratoga on
Pretty much everything on earth decodes MP3 and Vorbis via CPU.  DSP decoders offer no real advantage over CPUs for audio.  I doubt [random device] is any different then everything else.


I wasn't talking about everything else, I was talking about this specific device on which decoding Vorbis does indeed impact battery life severely, and has other unwanted consequences.  Playing back audio such as mp3 or aac which gets handled by the DSP sees the CPU useage between 5% and 10%.  Playing back Ogg Vorbis (using 3rd party ogg support package which adds tremor decoder to gstreamer) sees the CPU useage at minimum 40% and often going over 50%.  Naturally this also impacts on the ability to do other stuff at the same time.  It's a small tablet, not a single use music player, so typically I'd also have a www browser or a book reader open, or maybe a photo viewer or a map app  etc etc.  Multi tasking matters.  Nokia, being one of the owners of the aac and h264 patents, are a long time opponent of Ogg Vorbis and Theora for business reasons, and do their very best to make using Vorbis inconvenient and counter productive.  They succeeded in this respect.

FAAC vs LAME MP3 at about 128 kbps?

Reply #13 –
Quote from: Takla on
Wow.    That's a rather more aggressive tone than I'd ever expect in reply to such an innocuous post.  I'm sorry if my post irritated you but what to do?


I'm not irritated.  I'm just pointing out that since your decision was based on some irrelevant personal factors, its not a very helpful thing to share with others. 

Quote from: Takla on
Playing back audio such as mp3 or aac which gets handled by the DSP sees the CPU useage between 5% and 10%.


Mp3 decoding takes at most 25MHz on an ARM11 device, much less if you have a good decoder.  You've got a 400MHz CPU according to wikipedia.  25/400 = 6-7%.  Lets assume its in low power mode, which wikipedia says is as low as 165MHz. 25/165 = 15%.  If its using the DSP, it sure doesn't seem to do much of anything.  Actually, if it was using the DSP, shouldn't it be 0% CPU use?

It doesn't help much because its not using the DSP 

As I said above, people don't use DSP decoders for MP3 anymore because it makes no sense when you've got an ARM CPU.

Quote from: Takla on
Playing back Ogg Vorbis (using 3rd party ogg support package which adds tremor decoder to gstreamer) sees the CPU useage at minimum 40% and often going over 50%.


That sucks.  Is the device programmable?  Maybe someone will port a decent vorbis decoder.  There are very fast open source ones optimized for that CPU line.

FAAC vs LAME MP3 at about 128 kbps?

Reply #14 –
Actually the Nokia N series tablets do use the DSP to decode MP3.

Reference: http://maemo.org/development/documentation...a_architecture/


Code: [Select]
Format        Decoded with        Description
PCM         N/A     Raw PCM audio
MP2         DSP     MPEG audio layer-2
MP3         DSP     MPEG audio layer-3
AAC         DSP     Advanced Audio Coding, only LC and LTP profiles supported
AMR-NB         DSP     Adaptive Multi-Rate narrowband
AMR-WB         DSP     Adaptive Multi-Rate wideband
IMA ADPCM     CPU     Adaptive Differential Pulse Code Modulation
G.711 a-law     DSP     ITU-T standard for audio companding
G.711 mu-law     DSP     ITU-T standard for audio companding
WAV - MP3     DSP     MP3 audio in WAV container
WAV - PCM     DSP     PCM audio in WAV container
RM - RA10     DSP     RealAudio in RealMedia container. Uses closed source software, no support in GStreamer.


Even if mp3 is efficient enough that it can be decoded quite efficiently on the CPU it's still a much better idea to offload it to the DSP in this case.  This is a device that offers an environment not far away from a regular desktop, where the user expects to be able to run several quite demanding services and applications simultaneously.  One thing Nokia have done extremely well is power management, which clearly is an important factor in making this kind of choice.  If you follow the link above you'll see the multimedia architecture of these devices and when considering this combined with an overview of the devices' functionality and purpose perhaps it will make more sense.  The issue of ogg vorbis support is a longstanding 'won't fix' from Nokia because Vorbis is a direct competitor to them as patent holders in aac, h264 and mp3.  Unfortunately playback of ogg vorbis causes the media server daemon to place an extremely heavy load on the CPU. This pretty much kills the utility of the device, as well as eating the battery in double quick time.  Offloading the decoding to the DSP would be the only way to solve this but it is disallowed.  The devices run a Nokia OS based on Debian but heavily modified by Nokia.  There are proprietary drivers and plenty of the userland is proprietary as well, and the DSP is not supported by any driver other than the binary only propietary one.  Nokia have managed to take free software and encumber it (and lock out the users and developers from those parts it is extremely keen to remain untouched) to the extent that so far nobody has completely succeeded in replacing the OS or even been able to address this and similar issues in a really satisfactory manner.

Anyway I posted originally to offer my opinion and experience on the issue of low bitrate encoding.  It was something I'd not done before, generally preferring high quality Vorbis or MP3 for lossy.  So I did the boring thing, put my preconceptions aside, and tried out different encoders and was surprised.  That to me seemed relevant.  I put my opinions in a clear context so that the reader can decide for themselves their relevance or otherwise.

FAAC vs LAME MP3 at about 128 kbps?

Reply #15 –
This phone ought to support HE-AAC2 and then this very format is the best choice for you.