Using MP3 encoders

I encoded some language-instruction CDs today so that I could practice (or at least listen) while out and about, and found that the LAME encoder does a pretty clear job with speech. However, the files came out large, in part because LAME’s quality system for VBR sets an artificial floor on the bitrate – if you want the highest quality, everything gets a minimum of 128Kbps. Even the silences when I’m supposed to be repeating what the instructor said.

I tried fiddling with the quality setting, but it’s a simplistic 0…9 range. A lower setting reduces the voice quality too, not just the “silence quality”.

Does anyone know of a way to make the LAME encoder, or any other for that matter, use high bitrates only when necessary? If the pauses could be encoded at 32Kbps while the actual speaking parts got a much higher bitrate, the overall quality would be better and the files would be smaller.

MaPo,

Sounds like a limitation of the CD Ripping program you use. I usually use CDex and it doesn’t have this problem. You can set all kinds of options for the MP3 bit rates. It also uses LAME as the default encoder, though it can use others.

One thing you might also consider is setting the source to mono. Most instructional CDs are not stereo and you’re just wasting bits encoding in stereo. Personally the instructional stuff I’ve encoded was done in mono at a fixed bit rate of 96kbps and turned out just fine. I probably coulda gone down to 64kbps, but space wasn’t the primary concern.

Link: CDex

P.S. Which CDs are you studying and are they any good?

BTW, lest there be confusion – I’m using a VBR setting, not a fixed bitrate. Problem is, it’d be nice to have the voices at a high rate, while the silences are at low. At high (0) quality, LAME encodes it all between 128-160-192. At lower quality settings, the silences get bitrates as low as 32, but the voices also drop down into the 48-64-96 range.

I’d ripped to .wav separately. My soon-to-be-ex-computer is a frankencomputer at this point. I have to switch drives to boot Win95 in order to rip, Win98 to burn a CD, and boot Knoppix from the CD drive (which is then locked, so no ripping is possible) in order to encode. The .wav portion went just fine, nice and clear; encoding using LAME 3.95 goes ok, but I’m concerned about the clarity of the encodings in the end.

Still, it’s a LOT better than the last encoder I tried (RealNetwork’s crap), which had output that sounded like the speakers were living in a washing machine no matter what bitrate I used.

Hmm. If it’s using LAME as a back-end, though, it must either be a question of the settings you are using or of the flags it is passing. I’ll see if I can dig that out of the code (I’m guessing it’s a Linux prog, since it’s on SourceForge).

Oops! Great idea! That’ll do the trick all by itself, I think. I’d like to get them all on one MP3-CD, or better yet one (non-gigabyte) memory card so I can buy a Rio.

Pimsleur – they get rave reviews from most people. I used their Russian CDs a few years ago, and they were pretty good overall. My one quibble with them is that they have practically zero written material – the lessons are audio-only, so you don’t get any practice with cyrillic (or in this case, with pinyin or even pictographs). The whole booklet has only about half a dozen Chinese characters in it, and the few pinyinized words/phrases don’t have tones marked.

With Russian, I could at least use a dictionary to create written phrasebooks from the lessons, but it’s a lot harder with Chinese. . . .