Optimal mp3 quality/filesize for voice recordings?

[tranglos]

Looking for advice this time, not software

I need to prepare mp3s of speeches/presentations, recorded to wave with a high quality voice recorder. No matter what mp3 compression app I use, I'm finding that the lowest acceptable bitrate is 160 kbps. When I go below 160 kbps however, I'm getting a lot of artifacts in higher frequencies, the characteristic, infamous "underwater voice" distortion sound in mp3. The problem is that using 160 kbps yields files that are way too large for easy downloading (over 70 MB for a 60 minute recording).

At the same time I download a lot of podcasts which do not exhibit these artifacts, even though the bitrate used is 64 or even 32 kbps, and they have perfectly acceptable sound quality and definition. They typically fit 1 hour of speech in 20 MB or less. So what's the secret?

I'm guessing I should preprocess the wave file somehow, maybe use EQ to drop the high and low frequencies? Does anyone know what's the best way to go, or maybe there's a primer/how to for tnat sort of thing?

[MilesAhead]

You might ask Paul Thurrott as he does a lot of podcasting. Probably at least knows the basics of how they clean up the signal:

http://www.winsupersite.com/Default.asp

Or maybe Leo Laporte on Twit TV:
http://live.twit.tv/

Also this board subforum may be good:
http://forum.videohe....com/forums/33-Audio

edit: I'm not very experienced in audio but just from making expendable discs to play in my car stereo it seems .wma has better sound quality for music generally, at lower bit rates. I think the highest my ripper program would go is 192 .wma and it sounded as good or better than .mp3 at 320.  Of course that's music, not talking audio.  If you can use .wma you may try it as it might save you processing. It should inherently use smaller file sizes I would think.

[Shades]

As far as I know the general human voice uses the 3KHz band as a carrier. Which is why VPI and VCI settings from adsl modems  start with 3x (to mark the 3000Hz band). If I remenber correctly that bit of info was supplied to me while in the being in training for the communications section of the Dutch army.

Trimming/cutting away the rest of frequency bands in the wav file should already reduce the size of them significantly. 

[JavaJones]

What compressors have you tried? LAME should do quite well down to 32kbit/s with voice files. If MP3 is not a requirement,  try AAC. Higher quality for equivalent bitrate.

- Oshyan

[MilesAhead]

Sounds almost like he's having some kind of clipping problem.

I think I might try scatter shot approach and just pick 3 free audio tools at random from Softpedia and see if any produce clear output.

[tranglos]

Sounds almost like he's having some kind of clipping problem.
I think I might try scatter shot approach and just pick 3 free audio tools at random from Softpedia and see if any produce clear output.

-MilesAhead (June 25, 2010, 05:42 PM)

It's not clipping. The initial volume I get is actually pretty low, so I have to bump it, but that's not a problem. The problem is with typical mp3 artefacts that give you the impression of the voice coming from underwater, as well as "fuzzy", kind of rustling higher frequencies. You'll hear both in any music compressed with a low bitrate, 96k or less. It's just that I'm hearing this unpleasant effect at any bitrate below 160k, while it's absent from typical podcast recordings done at 64k.

It's nothing that would make the recordings unlistenable, but since I'm already doing it, I want to get it right, and keep the file size small, too.

I'm using RazorLame, which is LAME, or Sony Sound Forge, with very similar results. This is why I suppose I should preprocess the wave files somehow before compressing them to mp3.

[tranglos]

As far as I know the general human voice uses the 3KHz band as a carrier. Which is why VPI and VCI settings from adsl modems  start with 3x (to mark the 3000Hz band). If I remenber correctly that bit of info was supplied to me while in the being in training for the communications section of the Dutch army.

Trimming/cutting away the rest of frequency bands in the wav file should already reduce the size of them significantly. 

-Shades (June 20, 2010, 06:49 PM)

That makes sense, thanks, Shades. I'll be making new recordings over the next two days and will try your advice when processing them.