Hello,

I have never loved Microsoft, but recently I was impressed by the famous (or infamous?) 64 kbps WMA files, which sound good to me, for medium or almost-hi-fi listening (such as or good car stereo and mp3/wma portable readers).

However, I am able to hear that 96 kbps (CBR) WMA's are better than 64 kbps. After doing many tests by my own, I think I will encode all my wav files into WMA (VBR) quality 50. The bitrate of the created files are around 80-90 kbps: I am impressed with the relatively high cutoff frequency (> 12 kHz), without audible artifacts.

However, I found that MP3 files (produced with LAME) may be as good as those WMA's (between 64 and 96 kbps), if created with proper settings. I became aware of this by reading this old page (late 2002, when WMA 9 was released): http://www.mp3-tech.org/tests/wma9/index.html (by the way, the comparison he makes seems unfair to me, since he compares *CBR* WMA's to *ABR* MP3's!).

However, I am not sure how to easily get a simple (and possibly free) software that allows me to make those high quality mp3 files, as easily as I can do with WMA files. I have some mp3 software using Lame (up to 3.91) but they don't allow those lines with all those parameters.

What can I do?
I am ready to convert all my wav files to VBR WMA Quality 50 (around 80-90 kpbs), but... Is it good? Or is there a *simple* way to create MP3's as good as those WMA's (or better) at low bitrates?

Thanks
ipotesi

I think AAC-HE will provide better quality at that bitrate.

I do think that HE-AAC would be better as well. My own blind ABX tests suggest that HE-AAC is the best format for me when listening to low bitrate files (below 96kbps). However, there is plenty of free software that will allow you to rip your audio CDs using appropriate Lame settings. Both EAC and foobar2000 will do this. foobar2000 also has the ability to easily encode your lossless tracks using Lame.

Just make sure that you are conducting actual blind ABX tests and not just switching tracks on your computer, portable player, or car. A blind ABX test limits the placebo affect and is really the only advised method of testing here on hydrogenaudio. There are also some variations of ABX tests if you want to directly compare two lossy files.

It is generally regarded that portable environments (such as listening to an iPod and walking down the street or listening to music in your car regardless of the sound system) don't warrant the use of a high bitrate. Portable environments often mask certain frequencies by outside noise influences. Additionally, cars are not acoustically tuned and there are many outside influences such as the noise from the air conditioner/heater, engine, road noise, wind resistance, other cars, etc. The positioning of the speakers in the car also play a big role. You can move your speakers by less than 1 inch and that can affect the perceived sound quality.

So conduct your blind ABX tests with that in mind. You can use foobar2000 to convert your lossless files or rip your CDs and you can use EAC to rip your CDs. Just make sure that you follow the proper -V parameters when using Lame 3.98.2

For that given bitrate 80-90 and cutoff >12kHz, you could try listening a LAME -V8 or -V7 mp3 on your portable.

ADD: I tried to go up starting at -V9 up to -V5 using one very familiar track on my Nokia N70, played the tracks on random and not looking at the track name (ex. "Song -V9"), until I first heard an annoying encode, which really turns out -V9. -V8 and up goes fine on my phones headset, hence I use -V8 eversince. To think that it has 12500 lowpass, it still sounds fine. Maybe my earphones just suck, but hey, it's a blessing in disguise!

If you are going to use LAME then use the latest version (3.98) not the 3.91. If you want a GUI you can use foobar2000.

Get foobar2000, LAME 3.98, and then do some ABX tests, you can't go wrong with that. If you really need very low bitrates use V8 or V7, but I would suggest you perform your own ABX tests to find the best size/quality ratio.

Cheers.

This point is implicit in what others have said, but to get it up front:

you don't need "all those parameters." LAME has been developed so that you get optimal results with just the Vx parameters. Except in very special circumstances, more on the command line means less in sound quality.

(So, at least, people I trust here repeatedly say. Mindful of TOS 8, I have to say that I have not tested this for myself, but other people with better ears have.)

http://www.hydrogenaudio.org/forums/index....st&p=582652

(just being a sh*t disturber )

Thanks everybody! I did not expect to receive so many and good answers.

I have already downloaded Lame 3.98.2 (compiled for Windows) and I have created a few test files (running command lines), starting with VBR and CBR. I will download the suggested foobar2000 soon.

About VBR, I have tried -V 9, 8 and 7, with a couple of songs. The result is that -V 9 gives a very low average bitrate, around 40 kbps, while -V 8 gives more than 100 kbps, a big difference! I imagine there will not be that big difference all the times (for other files). However, I am considering to use ABR rather than VBR, so that I can decide the average bitrate: I hope that the loss of quality from VBR to ABR is minimal.

Among "all those parameters", the low pass frequency is the important one to my ears (with medium or almost-hi-fi devices). I tested that my ears can hear pure sounds up to 16 kHz, but for listening music I am happy with a cutoff frequency as low as 12.5 kHz. From that point, if I lower the cutoff frequency, the perceived quality rapidly becomes worse:
11 kHz: small difference, almost acceptable.
10 kHz: bad.
9 kHz: very bad.
8 kHz: horrible.

To ggf31416 and kornchild2002:
I have heard that AAC is better, but I am just considering the current devices I have now: they can only play MP3's and WMA's. At this time I am just trying to maximize the number of files for those devices, with minimal quality loss. That's why I will experiment with Lame instead of switching to WMA.
For example, I would like to have in my car just one CD-ROM for all the Progressive Rock that I like (total around 19.5 hours): this does not fit in the usual 700 MB CD with the standard 128 kbps MP3's, but it would fit if the average bitrate was around 90 kbps. The same for Classic Music + Excerpt from Operas that I like. Strange coincidence, this is about 19 hours also, and the same happens other four genres - generic rock, generic pop, Italian, and jazz: each is between 17 and 20 hours (so I might get everything I like into just 6 CD-ROM's, and fit it in my smartphone HTC Diamond with 4 GB and no memory expansion slot).

Thanks again,
ipotesi

You can now use floating point numbers, e.g.: -V8.5.

While this is all and good, just a little side warning. Anything less than -V5 will downsample to 32khz, and even further if you choose something like -V7 or -V8. Make sure your car can play these properly before wasting a lot of time making tiny mp3 files. Yes, believe me, I've seen things that can't decode anything but 44khz properly.

Really? The WIKI states no resample down to -V6 inclusive then a resample to 32kHz from -V7 to -V9 inclusive. Is it wrong?

Cheers, Slipstreem.

A quick test suggests the following:

Thanks for that. I've now read the LAME WIKI again (I may have misread it the first time) and it says 24kHz for -V9. Could someone edit it please?

Cheers, Slipstreem.

Hi ipotesi, I see that you are concertrating your attention on file size (or bit rate), but did you also listen to the results at V7 V8 V9 etc and determine what is acceptable to you in terms of quality? I wouldn't get too carried away with only filesize and ignore quality.

I don't usually use quality settings below V5 but I just did some tests at V8 to V9 to see what it bit rate I got. I also found that there was a big drop in bit-rate (and also a huge drop in quality) in going from V8 down to V9. Here are the results for the average bitrate with the one complete album that I tested :

V8 - 94kbps
V9 - 64kbps
V8.1 - 89kbps

As for the quality I've got to say that I was quite surprised with just how good V8 sounded! I'll blame it on the cold that I'm currently suffering from, but I was having some serious difficulty ABX'ing these V8 mp3's from my lossless originals. Ouch I didn't expect that!

Now V9 on the other hand was a very different story. It was very noticably poor quailty and trivial to ABX even in my current state. It seems that each increment quality setting from V5 down to V8 only gives fairly incremental changes in bitrate and also in quality. The change from V8 to V9 however seemed almost like a total collapse in quailty rather than an incremental change.

Personally I'd recommend sticking with true VBR, and in particular using V8 if you can. If you must get to 90kbps or less then try the new "floating point" settings like the V8.1 setting I tested above. I'd stay as close as possible to 8 if I were you, it's got the 12.5Khz cutoff that you dont want to go below and sounds absolutely great to me. I might even switch to V8 for my (low quaility) portable player.

Regarding VBR versus ABR. Yes it's true that the av biterate with VBR does vary quite a lot from song to song and from album to album, however the average over a large number of albums is much more predictable. Since you're really only interested in what you can fit into 700MB then you're really only interested in the average over a large number anyway, so why sacrifice the extra quality of VBR when you can fine tune the average with the floating point quality settings.

Instead of ABX tests, ABC/HR tests would be much more useful in this situation.

http://wiki.hydrogenaudio.org/index.php?title=ABC/HR

The problem with VBR, even with floating points such as -V 8.5 , is that a big leap remains:
there is a crash in quality between -V 8.4 (bitrate around 88 kbps, automatically resampled 32 kHz)
and -V 8.5 (bitrate around 42 kbps! Automatically resampled 24 kHz).

By the way, if I make a -V8 or -V9 with --resample 44 and/or --lowpass 12 or 13, the bitrate skyrockets to >110 kbps (it seems to become a -V 6 file, even if I want to make it as a -V 9 or -V 8 file).

That's why I think I will use ABR around 80 or 90 kpbs: I can control it much more than VBR, that works too "automatically".
Or I will switch to WMA VBR (which I would not like to do, because MP3 is more "universal" than WMA, and I have never loved Bill Gates anyway).

ipotesi
Italy

For mp3 at 96kbs, try either of these options:

1) get Lame 3.93.1 and use this cmd line: --alt-preset cbr 96 -h --resample 44 --lowpass 16 (OR) --lowpass 15

2) get Cool Edit Pro v2.0 and in the mp3 settings select 96kbps 44100Hz Stereo, 16000Hz or 15000Hz lowpass (bandwidth), Fast Codec and uncheck error correction checksums.

Don't ask why, just try it. It is the result of long-term encoder testing at this bitrate.

Isn't that an illogical statement? you aren't happy with the decimal values of -V setting in lame, and think about using an encoder that only offers five levels?

I ABR suits your needs better, use ABR, but don't pretend to make the false assumption that the bitrate is an absolute measure of the quality of an encoding.
(Of course, the VBR algorithm isn't tuned to be used at those low settings)

It depends.

If one ignores the real purpose of the poster, that statement may seem illogical (as the other posts of the same poster).

From the standpoint of the poster, who happens to have his collection separated by genres, with each genre occupying around 900-1000 MB at the usual supposed standard (128 kbps CBR) of mp3 files (that means 1 CD + 30% for each genre), it is logical to reduce the file length of 30%, just to have a single CD for each genre, while trying to save the maximum possible quality in doing that.

Unfortunately, it happens that the good VBR settings start at an average bitrate (115 kbps) that is just a little too much, and exactly destroys the purpose of the poster, just for a minimal exceeding amount (it does not make sense for me to reduce each genre, from 1 CD + 30%, to 1 CD + a little bit exceeding the space of that single CD!). Both Lame mp3 -V 6, and WMA Standard 9.2 VBR Quality 50, are around 115 kbps (average). So, I am trying to find out how to reduce the bitrate just a little bit to make each single genre fit in a single CD. But there is a leap that causes the bitrate to drop too much (from -V6 to -V7, or to WMA Quality 50 to 75) for the purpose of the poster. So, he is trying to find out if a good solution is possible, possibly around 90-100 kbps.



As I pointed out, the decimal values of the -V setting does not eliminate the huge gap that remains between -V 8.4 and 8.5, so the decimal feature seems misleading to the poster, who is considering to use the WMA VBR Quality 50 since it seems to fit better the bitrate needs of the poster. But even in that case, there is a leap between 80 kbps (Quality 50) and 115 or 120 kbps (Quality 75).



I am only trying to find a bitrate that solves the above problem: it would be around 90-100 kbps. And it *is* actually a matter of bitrate, just for the nature of the problem itself.

Thanks for your help,

Ipotesi
(or, in Catalan, Hipotesi)

ipotesi, have you tried lame at around -V8.0 to -V8.2, it should easily get you're target bit rate when averaged over a larg number of albums?

Do you have some predisposition against 32kHz sample rate? 32kHz can easily handle the 12.5kHz bandwidth that this quality requires so there is no real downside. Have you done any careful listening at -V8 yet?

To uart:
Probably you are right when you say that I am misleaded by the 32 kHz resampling: psychologically I would prefer 44.1 kHz, even if there is not a logical reason.
I am now considering --ABR 104 (Lame 3.98.1), which automatically creates 44 kHz files with a cutoff frequency just above 15 kHz. The average bitrate is higher than my original purposes (80-96k), but it seems the best solution to me at this time. I expect it to be better than -V 8 (which would produce overall a lower average bitrate, around 85 k).

To jmartis:

Thanks for the suggestion: I have downloaded Lame 3.93.1 but... I have to ask why!
First of all, why an older version? (I read somewhere that 3.97 and 3.98 may produce worse files than previous versions: is it true?!)
Second, why CBR rather than ABR? This is really surprising to me.
Finally, about the lowpass, I think that 15 would better than 16, to avoid more artifacts.

Thanks,
Ipotesi

All this effort to save a 10 pence CDR?

Mad!

btw, I hope you're not using those 128kbps CBR mp3 files as the source for this test. People sometimes get a bit too hung up on transcoding and how bad it can be, but 128kbps CBR mp3 is not a good source!


I guess you're not planning to buy any more music, either?!

Cheers,
David.


In theory (and sometimes in practice) 44.1kHz suffers from mp3 pre-echo less than 32kHz (fixed frame size in samples, so pre-echo extends further in 32kHz version).

However, there are simply more samples to store at 44.1kHz than 32kHz. A low pass filter reduces the difference, but does not eliminate it. 32kHz is more efficient - usually more than enough to counteract the possible per-echo issue.

FWIW some people encode to 32kHz even at 128kbps. The only downside is compatibility, especially with CD burning apps if you want to write the file back to an audio (CDDA) CD - some will perform poor quality resampling back up to 44.1kHz. There must be some players somewhere that only work at 44.1kHz, though many PC sound cards are actually measurably happier with 32kHz (and 48kHz) than with 44.1kHz (you'd be hard pressed to actually hear the difference though).

Cheers,
David.

Yes I've got to admit I was thinking the same thing.



Yes but lets hope he doesn't plan to write low bitrate mp3's back to CDA. Better not to do this anyway.


Well I think it's time to get over preconceptions and get down to some serious listening tests!. Only then will you have some real basis for making a choice. Like I said before, I was surprised how good -V8.0 sounded.

Yes, at least according to my tests, the new Lame CBR and ABR settings produce much more varying result, where the older versions are more "stable".
You can try ABR too, but again, according to my long time tests, the CBR setting always produces good and solid results, I've yet to find a sample where my setting fails badly. I've not tested the ABR much, but it appears that it can produce better results on some samples and worse on others - in other words, it's not as stable.

Try my setting and tell me what you think. It will even cope with the 16kHz lowpass most of the time without audible ringing or other artifacts.

-J.M.


Edit: It can happen you'll rarely notice low frequency distortion/hiss, in this case adding "--ns-bass -4" switch should make it less audible, you also can include this setting all the time as it really does not decrease overall quality much.

I think 192 kbps is much more standard bitrate

Man, being a fan of progressive rock are you really willing to sacrifice quality just to squeeze everything in one cd??

If they can't properly ABX the difference between 80kbps mp3 files and the source lossless files then they really aren't sacrificing any quality. To them, the 80kbps mp3 files sound just like the lossless ones. Not everyone has the same equipment and ears. Additionally, cars are not the best environments for listening to music as there are just too many outside influences that affect the perceived quality of music.

Yes... My collection is classified in 7 genres:
Progressive Rock 1
Progressive Rock 2
General Rock
International Pop
Italian Pop
Disco/Dance
Classical/Operas

Getting the best from each genre (at 128 kbps CBR) I reach the capacity of one CD + a little bit more (approximately the same for each of the 7 genres!). So I keep 14 CD's in my car. With some more compression, they could be only 7, that would be more orderly and logical. Of course, I don't need a very high quality in my car (or in portable mp3/wma readers). I can listen to the CD's at home if I want high quality.

All started when I happened to listen to the famous (or infamous) WMA CBR 64k and I was impressed, since its quality seemed to be very good.
Then I made a few tests that showed that they were not so good, and I thought I could achieve better results with WMA CBR 80k or 96k. Then I discovered the WMA VBR Quality 50 option.
And finally I discovered that it is not necessary to switch to WMA's, because Lame is able to achieve good results around 85k-100k: so the matter became more and more complex and you know the rest of the story...

At this time I am considering several options, including:

WMA 9.2 Standard Quality 50 VBR.

The Lame VBR (-V 8) suggested by uart.

Lame ABR around 96k, that seems to be very good according to this listening test: http://forum.hardware.fr/hfr/VideoSon/Trai...jet_84950_1.htm (* see note).

The old-version Lame CBR 96k suggested by jmartis, even if I can't figure out how such settings can achieve a good quality and keeping it even with a 16 kHz cutoff!


I don't like the music that have been produced in the last 15 years or so: I am only buying one single every two or three years! The last album I bought was Images and Words by Dream Theater (1992), apart from a collection of Simply Red (Greatest Hits, 1998, including 1985-1997 songs).


I have tested that the 32 kHz resampling actually achieves slightly better results than the original 44.1 kHz (around 96 kbps).

Before doing that, I had the following idea to quickly verify whether the overall principle was right (that is, whether there is a loss of quality due to the attempt to keep an high sampling rate): I applied this principle to very low bitrates, such as 32 kbps CBR, which is automatically resampled by Lame at 16 kHz only (with a cutoff around 7 kHz).
Then, starting again from the original wav file, and keeping the 32 kbps setting, I added the parameter --resample 44. The file size was the same, but the quality was much worse, with extremely horrible artifacts, that is what I expected due to the huge difference in sampling frequencies (44.1 and 16 kHz). The artifacts were not due to the cutoff frequency, which remained the same (around 7 kHz).
Strong artifacts can be heard (with 32 kbps CBR) also resampling at 32 kHz, and even at 24 kHz and 22 kHz (there is no need to make ABX tests in these cases, due to the strong difference).

Ipotesi

(*) According to the listening tests I read, he Lame quality strongly increases from 80 kbps (worse than all the competitors) to 96 kbps (among the best codecs!). Probably the 32 kHz sampling helps to get a better quality than competitors, since it needs to compute fewer bits than at 44.1 kHz. See http://forum.hardware.fr/hfr/VideoSon/Trai...jet_84950_1.htm (in French)

For prog rock you are going to have problems maybe not today but sooner or later abx or not and end up reencoding everything . But with no regards for quality why not ? go for lowest bitrate .

You need to consider song intros , solo instruments, low volume vocals as prog has plenty and they will sound like utter cr-p esp at night or earphones. All subjective though but i've been there and ran away fast.