GoldWave Inc.

I am not well-versed on the different file format attributes for mp3. If I am recording and saving a radio show with no music, what would be the best file attribute (i.e. Layer-x, xxxHz, xxx kps, stereo/mon) to give me the smallest file and not reduce the quality too much?

Thanks.

I can't directly answer your question, but I can give you some theory that may help you to tweak your settings. (I don't use MP3 that much, and when I do, I'm using it for high quality music. So, I'll let someone else give you a starting-point.)

The file size is determined by the bitrate (kbps) = kilobits per second) and the playing time.

This formula can help you estimate file size:
File Size in MB = (Bitrate in kbps x Playing Time in minutes) / 139.8

For a given bitrate, you will get higher quality with mono, since the "bits per second" don't have to be allocated to two channels. (MP3 also has a "joint stereo" option, which makes better use of the limited data by not duplicating data in both channels.)

The sample rate (kHz) limits the high frequencies. (The file can't contain any frequency-information above half the sample-rate.) For voice, you can get-away with a 22.05kHz or less. For example, digital telephone systems use a sample rate of about 8khz. For voice, "T" and "S" sounds contain the most high-frequency information, so if you use too-low of a sample-rate, you start to loose those sounds. (A lower sample rate doesn't change the file size, but it forces the MP3 encoder to make "better use" of the data by ignoring the high-frequency details.)


P.S.
MP3 is always "layer-3".
And, you have a couple of other another options... variable bitrate and average bitrate both make MP3 "smarter" by using more bits where there is lots of audio detail, and fewer bits where where there is less detail.

I store a lot of voice-only as MP3 and based on no scientific theory or algorithms whatsoever, just based on what sounds "ok" to me and is fairly small, this is the setting I use:

Layer-3 ACM, 22050 Hz, 32 kbps, mono.

I think the ACM has something to do with the encoder - maybe that the Frauenhofer MP3 encoder is being used instead of the LAME? Not sure that's more one for the technos.

I initially encoded at 56 kbps but found that 32 kbps was satisfactory at almost half the eventual size. I did notice too much of a drop-off at 24 kbps though.

I would probably experiment with VBR if I had more time but for now I'm just using constant bitrate. Particularly with voice only it would seem VBR might make sense.

Hi,
I agree with Holtram in that 32kbps is quite adequate for voice or radio encoding. I have herds of old radio shows and the 32kbps setting gives good results.
Cheers,
Fred

I have to agree about the 32 kbps. It's very adequate. I myself completed a project in July where I took all 17 CD's from the last Harry Potter series and was able to "squeeze" them onto a single CD. I ended up using 64 kbps - and they "just fit". The advantage is that I don't have to fumble 17 times changing CD's in my car. So, Goldwave keeps me safe on the road!!

File Size in MB = (Bitrate in kbps x Playing Time in minutes) / 139.8

that works, but it's not totally accurate. back in 1999 when i took a VB class in high school...i wound up writing this program i wound up calling ACSC (Audio Compression Size Calculator), it was essentially a real simple VB program that took a could of text box inputs where you punched in easy-to-under stand numbers, picked a preset birate (or entered your own), and it'd not only calculate the total filesize, but it'd give you the compression ratio as well. there are a few reasons it doesn't exist today...largly because I've since lost all the original VB project and exe files...and the second release was highly polished and was a great improvement, but that's when VBR started being supported and i figured it was kind of a dead horse, so i abandoned it.

however, i at one point had a very overly complex formula that spit out some pretty accurate results....this is a smplified version:

(((bitratekbps x 1000) / 8) x seconds) / 1024

as far as sampling rate...that's half right, it's true it limits the high frequencies because of how audio is sampled...but mostly the sample rate determines how many samples a second are going to be used...it's just most people keep thinking bay to Nyquist and think the only thing that should determine the sample rate is your desired maximum frequency.....

that works, but it's not totally accurate.

You are correct. I said "estimate". There is some overhead in an MP3 file. It's usually not much unless you have embedded album artwork, etc. But, you can't be 100% accurate.

(((bitratekbps x 1000) / 8) x seconds) / 1024

HA! HA! That's EXACTLY the same formula! Except, I combined the 3 constants. I used minutes. And, your results are in kbytes (1024kb per Mb).

as far as sampling rate...that's half right, it's true it limits the high frequencies because of how audio is sampled...but mostly the sample rate determines how many samples a second are going to be used...

Huh? The sample rate is samples-per-second. It's effect is to limit high frequencies. (It has a direct effect on uncompressed file size.)

ok ok, i was getting different results because i forgot i had an additional / 1024 in my formula...of course that was if the output was over 1024 kbytes.

you say the sample rates job is to LIMIT the high frequencies...how many times to do i have to say it..technically...no..it's not. the fact it does is just a byproduct of how digital audio works. if i wanted to have a 96khz file that was band-limited to 22khz...i could have that....but that doesn't mean i HAVE to drop down to 44.1 khz to get that....why would i do that? i'd have over twice the information to create that 22khz sound......sample rate is almost kind of like tape speed..the faster (more samples) you run, the better/more accurate the digitization.....

what looks better....a connnect the dots with 44,100 dots, or the same picture using 96,000 dots? think about that and maybe you'll see where i'm coming from with this.

you say the sample rates job is to LIMIT the high frequencies...

I didn't say that. I was explaining to cakaplan how a low sample rate would affect the sound.