Hi,

Due to space constraints for a project I am doing I need to encode 64 kb/s mp3s. I will be using Lame. What settings should I use to get the best result possible in this low bitrate? It doesn't matter if takes long to encode since we're not talking very many track. I am going to encode from wav files.

//C

Better off with FhG encoder using IS, public listening tests are around....

The Lame setting that I would use is -V 9. That would still produce VBR files so the bitrate might be a little lower or a little higher than 64kbps. The next setting that I would use is --preset 64 (or --preset 70) which should produce ABR files at around 64kbps.

westgroveq is right though, you would be better off using FhG to obtain optimal quality as past listening tests have shown that it is better at the sub-96kbps bitrates. That is only if you want to achieve optimal quality for your project. Go ahead and use Lame if your project requires it or if your project is actually a listening test.

mono.

VBR if at all possible.

Cheers,
David.

Thanks guys.

The project is a presentation of sorts stored on a tiny USB memory stick. The size of those is already given (the memory sticks have been purchased), so I need to fit the embedded sound tracks within the space left. I can use mp3 or wma, but since I'm running Linux I can't encode wma far as I know. This is going to be played on laptops and desktop computers with low end and/or built in speakers, so sonic bliss is not crucial. However, I would like it to sound as good as possible of course.

I have had some 64 kb/s files on my portable player and in my stereo. They sounded good to me (probably because I didn't compare them to higher bitrate ones). I have never tried mp3s lower than 128 kb/s, but I would assume it's not totally different from the wma's?

One possibly stupid question: Is the FhG available for Linux and would I have to purchase it?

//C

Seconded on both counts. Go for mono, and use a VBR setting (-V 9).

Can you, please, stop recommending -V9?

From the lame documentation:

for this bitrate, --abr 56 -m m . And yes, that means mono.

I recommended -V 9 as I didn't know what the OP was encoding. Now that I see that they are encoding a presentation (ie voice), it would be alright to use ABR especially if you want to fit the files in a certain space. ABR will give you more of a known file size but this doesn't really matter for voice. Hell, I think even 64kbps CBR would be fine for voice.

The underlined sentence is actually a link.

Concretely, I wanted to point to this line -V 8 and -V 9 are out of the question for acceptable quality (i.e. not warbling, dropouts and other annoying artifacts)

I may not have been clear, but the sound tracks in this presentation is actually music, not voice. It's pictures and text with music. The music is kinda important, but isn't the main feature of this presentation.

//C

Try these settings and use the one that fits better (size/quality):

lame -V 6 -m m
lame -V 7 -m m
lame --abr 56 -m m
lame --abr 64 -m m

At this bitrate, and given your target usage, there's no good reason to use stereo.
It will sound good enough. I just wanted to say that -V 8 and -V 9 have never been recommended, and there's a reason for it.

[Edit: removed unnecessary quote]

@JAZ

May I know reasons why -V8 has never been recommended? I personally would recommend -V8 for portable listening if the case maybe (but not -V9).

However in the OP's case, it is not recommended, since -V8 stereo averages 80-90 kbps.

The technical reasons are due to the way the quality is scaled.

With lossy encoding in general, quality does not translate to change one or two settings gradually. It actually translates to do that *and* disable or switch the behaviour of some of the tools involved.

In doing so, the quality does not scale the same way, and the bitrate distribution may not represent the best accuracy possible.

Two examples:

Nero's MP4 encoder:
It switches from LC to HE to HEv2 depending on the quality scale. The enabling of these tools translate to a need to adapt the other parameters in a non-linear way. (And that's also a reason why using the switch -lc makes the quality scale different from not using it)

LAME:
It reduces the samplerate depending on the -V setting. This change not only reduces the bandwidth, but also the precision of the tools (It used to be said that LAME is tuned at 44Khz, not sure how much of that holds true nowadays).

But there's also some explanation in how a VBR file is created vs. how a CBR file is:

A VBR file will adjust mostly the allowed masking thresholds (giving more or less margin) to reduce the bitrate.
Other factors that are considered in LAME are the bandwidth (lowpass), the sfb21 limitation (-Y) and joint stereo switching threshold. (This list may not be accurate, but should give an idea)

A CBR file will loop between the ways to produce the expected bitrate, allocating the best compromise between the real and the wanted bitrate.
In some cases, it may need to do drastic solutions, and that's when artifacts show up (dropouts, stereo collapse.. )

But most probably, the real reason is simply that it has not been as finetuned as -V 2 or -V 5.


As to why specifically not -V 8? It's 32Khz like -V 7, you'd say, but compared side by side with an abr setting of the same average bitrage, the abr is better. Do a few tests..
[Edit: Note that --abr may be using a higher lowpass, so at your discretion, use the same one)

do you really really need to use mp3?