Export - smallest size, best quality

Hello, I’m trying to do an export of a large (50mb) mp3 and get the smallest file size with the best quality balance. Size is preferably 5 mb or less…I’m using Lame Library, I also tried OGG and I can’t seem to get less than 15 mb. Can anyone suggest anything?


Was the original sound file an MP3, or you just think it was? If you are starting out life with a compressed MP3, any further compression will make the show turn into old peanut butter in a big hurry. Each compression pass adds damage and you end up compressing the damage and not the show.

The only way to make these tools sparkle is to start out life with a perfect, large, uncompressed WAV or AIFF of the show. The higher quality the better. Then you can really use all the MP3 tools without significantly damaging the show. One thing you can do before the compression step is make the show mono instead of stereo if you haven’t done that already. Boom. There’s half the file size right there.

If you don’t have to have perfect interchangeability with other computers, Windows WMA and Mac QuickTime M4A both use the H.264 compression tools and they can run rings around the much older MP3. I’ve known web sites to post both with very good quality and impossibly small file sizes.


I thought this might be a faq, but I’m having trouble getting a sensible mp3 output. I have the latest LAME library (dll), but it just makes HUGE files!
I gave up generating mp3s with Audacity a while back, instead what I do is export to WAV and just use the command line to encode.
For example, I’ve just encoded a 28 minute file. With Audacity, variable, quality 6, it came out at 30Mb.
With the following command line option:
-h -V 8 --vbr-new --resample 22.05
I can get something which, for speech sounds to me (even via headphones) almost identical. Sure, the music at the bottom “tinkles” slightly, but it seem that if I use the v9 export option, although the file size is around the 9Mb mark, the quality is MUCH worse than the 7Mb “manually” encoded file.

So… I guess my question would be: Is there a way to add my own option to the export page?

As you no doubt found when you opened up the lame help pages, lame goes forever. True to its roots, “Lame Ain’t an MP3 Encoder,” you an use lame for thousands of different options and certainly not just making MP3. Audacity has a vanishingly small subset of options available because nobody wants the career move of mastering all of them.

Also, most people don’t want slightly bubbly voices. That’s what Sirius and XM radio sounds like and I can’t listen to them. Most people want perfect-sounding musical performances and complain bitterly when they don’t get them.

Correction, most people want perfect musical performances in 2KB. That means they don’t want MP3. On a good day, MPEG 1, Layer III, designed in the twelfth century, can’t do that.

And yes, certainly, nobody is stopping you from running lame outside of Audacity.

-h -V 8 --vbr-new --resample 22.05

I can just follow that. High Quality, Voice??, 8 bit (mono??), Missing Default Option, Variable Bit Rate…etc…etc. It wouldn’t surprise me if you went to all those options, mono and a much lower encode rate, that the file size would drop a lot. I’m also betting that if you started with an MP3 instead of a WAV file, those settings would produce gargly trash

You win.


OK, you’ve lost me there! I can’t see any other format encoding options, and Wikipedia says:

“LAME is an open source application used to encode audio into the MP3 file format. As recent LAME releases are no longer a patch against ISO encoder code, LAME is now itself an MP3 encoder; the LAME acronym has become a misnomer.”

Well, what CAN do that and, more importantly, be streamed over the now-ubiquitous variety of flash-based players out there? ogg can shave a couple of k off for the same quality, but realistically, I want something people can use. Same goes for AAC - nice codec, but where’s the web support?

I just experimented using the settings guide at Hydrogenaudio
-h = high quality
-v = variable bitrate
8 = quality of vbr
–vbr-new = type of vbr encoding
–resample 22.05 = well, for flash players, it’s got to be a multiple off 11.025. 44 seems wasted on voice, 11 is too low, so 22 it is.

How do you mean “missing default option”?
Well, let’s do an experiment!

Here’s a wav of the “original” file - remember, this is an “imperfect source” that’s already been encoded as realaudio, so this is the kind of file I’m dealing with.
http://www.digitaltoast.co.uk/audacity_samples/stereo_wav.wav (27 seconds, 4mb)

Here’s audacity’s output at various settings (I’m sure you can work it out from the filename - mouseover and watch the browser status bar as the forum seems to have truncated the names)
http://www.digitaltoast.co.uk/audacity_samples/stereo_mp3_audacity_vbr_fast_q7.mp3 < 466KB

http://www.digitaltoast.co.uk/audacity_samples/stereo_mp3_audacity_vbr_fast_q8.mp3 < 394Kb

http://www.digitaltoast.co.uk/audacity_samples/stereo_mp3_audacity_vbr_fast_q9.mp3 < 331Kb

http://www.digitaltoast.co.uk/audacity_samples/mono_mp3_audacity_vbr_fast_q9.mp3 < 168Kb

Now for some “directly encoded” LAME files (or at least using all2lame with settings of -h -V 8 --vbr-new --resample 22.05)

http://www.digitaltoast.co.uk/audacity_samples/stereo_all2lame22_05.mp3 < 226Kb

http://www.digitaltoast.co.uk/audacity_samples/mono_all2lame_22_05.mp3 < 134Kb

To me, via desktop speakers, they all sound the same. Via good quality Sennheiser monitor headphones, the difference is just discernible.

(I notice now is that Audacity isn’t trying to encode as 32Khz which is a good thing!)

Well, I’m starting with a realplayer stream, opened using the ffmpeg decoder in Audacity.
Let’s so some examples: Here’s a raw wav of the real stream, so you know how is sounds to begin with. It’s a bit of speech but with other background noises so it’s complicated.

?!? I win what?

I see what they did. lame creates “Compressed Audio Files” that can be read by most MP1, 2, and 3 players.

On the linux machines here, if you do a ‘man lame’ to get the instructions, you get 836 lines – several pages – of options and switches. Not for faint of heart, this. There certainly are defaults on a lot of the options and Audacity cherry-picked from those. Did I see one of the setups give you the ability to create WAV files?

<<<-h = high quality >>>

I’m doing that entirely from all too fuzzy memory. Generally, when you have two dashes, it means you want one of the options in the middle to come out a default number. You need the extra dash to signify you decline to set one value.

<<<?!? I win what?>>>

Any time you create a desirable result, you win. I suppose if you created a result desirable to both you and the client, you could conceivably also win a tray of warm brownies.

It does remain, however, that if you compress an already compressed performance, you will be compressing and managing the errors in addition to the show. All but a few very gentle compression schemes create distortion and errors.

H.264 variations; WMA, M4A, AAC, etc. are rapidly taking over distribution because of their ability to pack quality into a really tiny digital package. If you have a Windows machine, you have Windows Media. If you have an iPod/iTunes/Apple, you have AAC. I thought the latest FireFox knew what M4A was, but you still need to add software.

Oh, well.


If you change the project rate (in Project Rate box in the lower left of the main Audacity screen) to a lower rate before exporting, the encoded MP3 will use that sample rate and produce a smaller file.

Reducing the sample rate will also reduce the frequency band limit (the maximum frequency at a given sample rate is always less that half of the sample rate). If you are only interested in the intelligibility of the voices, you can probably go down to around 16kHz sample rate (which will limit the upper frequency response to around 7kHz) and still have acceptable quality.

Converting to mono will make a big saving in “file size : quality”.

To my ear, Ogg performs better than MP3 at high compression settings.

For speech only, “speex” format can achieve remarkably high compression (not relevant for the audio sample that you posted, but may be of interest).