I have a LOT of files that are to be converted from CDA to MP3. They are voice-only files. The ideal target bit rate is 20K to 32K. I'm keenly aware of lower bit rates having lower quality, but what I don't understand is why WMA and other formats sound so much better at 32K than MP3 does. (I'm not talking subtle differences between encoders.)

If this topic is of interest to you, listen to any WMA file here. Notice that the stream is at 32K stereo. Compare that with the 40K stream here. The 32K stereo stream is superior to the 40K mono stream, particularly if you listen on headphones. Both sources use professional grade equipment. The source audio files are comparable.

So the question is, what is the secreat to audio processing prior to conversion that can yield better MP3 files?

If you are going to convert voice-only files, consider using a speech codec. I don't have experience with any of those speech codecs out there, but I think it'll suite your need.

About the different of sound quality between codecs, it's easy to explain. MP3 or WMA or AAC or a lot more (except those lossless codecs) are all lossy compression.

It mean, the codec have to strip some information from the source to keep the size of compressed data small. If a codec have good algorythm to compress the sound data, its compressed file will still keep good sound even with lower bitrate.

Mp3 for the most part does not encode at low bitrates as well as newer codecs such as Ogg Vorbis. Mp3 starts sounding almost decent, for my tastes, at 96 kbps and up. Your miles may vary.

If your are are restricted to encoding to the mp3 format, I would recommend using the latest 3.97 alpha version which you can obtain from, among other places, www.rarewares.org. If you prefer staying away from an alpha release even though quite a lot of testing has already been done to show 3.97 is quite respectable, you may choose a stable release such as 3.90.2 or 3.90.3.

Try encoding at this setting:

--preset cbr 32 -a

The "a" parameter encodes to mono, but since your sources are voice files, this is a good tradeoff to obtain "good" sounding mp3's at that bitrate.

There is no reason to use 3.90.2.

Use 3.90.3 or 3.96.1 if you're going to use a stable version.

Why not use HE-AAC? That's how I do my recorded conferences.

Correct. "Speex" at around 28kbit average bitrate sounds quite awesome with voice-only content.

There's a number of reasons for this. The first is LAME isn't really tuned for bit rates that low. LAME users tend to prefer transparent or near transparent quality at higher bitrates as opposed to acceptable quality very low bit rates.

Secondly, WMA and Vorbis as formats are technically more capable of handling lower bit rates than MP3 simply due to the inherent nature of the codec's design (tuning on the encoder side also helps).

Thirdly, a number of codecs (such as HE-AAC) use specialised techniques (spectral band replication and parametric stereo for example) to achieve better results at low bit rates, and as a result will typically sound better than MP3. These cannot be used for MP3 because the MP3 standard is already defined (Mp3 Pro, a newer standard, does use SBR).

I'm getting better results by converting from CDA to RA and then from RA to MP3. The only thing I can figure is that RA's pre-processing must be better than LAME's. Very strange.

LOL ... that's hilarious! And here I thought you were serious with your first post. Great joke! You must have quite a sense of humor.

Regards,
Madrigal

This could be true. Real is usually applying heavy preprocessing for very low bitrates, while Lame is only lowpassing.

If preprocessing is the key, then maybe you could get better results by using a *real* voice codec, (pun unintended) like Speex.

If you don't absolutely need MP3 files, don't transcode from a voice codec to MP3. Just use the voice files. And if the result doesn't need to be stereo, you can downmix the source to mono for some extra compression.