Which codec (mp3,WMA, or AAC) is better for just voices?

[nstgc]

As we all know (or should) mp3 is pretty much the worst of the popular lossy codecs or listening to music. However, I assume its because it was never meant to be used for music or other back ground sounds. So, I thought I should ask.

I have a bunch of listen CDs for learning German. I was wondering if anyone knew which would be best.

[edit] These are to be placed on my DAP, which can play AAC, WMA, and MP3 without converting. Also, FLAC, or any other lossless would be overkill. I don't intend on using even 320kbps when encoding.

[kitamesume]

o.o although 320kbps mp3s are pretty good, around 80-90% near that of a good enough flac record.

though why not flac then? they arent as huge as wav =P

edit: what are you up to anyway, if its for saving space then 320kbps mp3 is already overkill for voice records, musics has their range because of other instruments but voices arent even outside the 80hz-1.1khz range, although in extreme cases they could reach 60hz-7khz range, still not enough to reach the maximum 20hz-20khz recording bracket.
edit2: that was singing btw, speech only has a range of 80hz-255hz.

[datora]

.
I assume your goal is to have spoken files that are clear enough to learn from, and you wish to save filespace to a 'large degree.'

In which case I would go with *.mp3s for broad & simple compatibility.  128 Kb is overkill for voice only.  I generally don't rip in anything less than 192, even voice, but I do it knowing that I can't hear the difference.  It's a 'comfort zone' thing that I know the quality exceeds my hearing, and the minor extra space used is unnoticeable.  I think the next usual step-down os 96(?) Kb.  Which should be fine for voice-only if you're really pressed for space.

You might want to just take a couple of short sample files, like about a minute each, and code them to different formats & different compression levels.  Put on a solid pair of headphones and listen carefully to determine for yourself what compromises best work for you.

If it was me, I'd rip it all to 192 *.mp3 and not think about it further.  Also, consider variable bit rate and set the range from 96 to 192.  You'll get smaller files than CBR 128 and all the important parts will be at 192 clarity.  I've heard files in the past set to VBR 60-128 and they were quite passable for basic use ... but, that wasn't for teaching myself a spoken language.

[megido-rev.M]

For simply voices, just use MP3 VBR with a low lower bound.

[vuzedome]

datora has my thoughts, mp3 at 192kbps will do the job quite nicely.

[Pentium100]

edit: what are you up to anyway, if its for saving space then 320kbps mp3 is already overkill for voice records, musics has their range because of other instruments but voices arent even outside the 80hz-1.1khz range, although in extreme cases they could reach 60hz-7khz range, still not enough to reach the maximum 20hz-20khz recording bracket.
edit2: that was singing btw, speech only has a range of 80hz-255hz.

Voice goes up to about 8kHz - while the fundamental frequency does not go that high, the harmonics do. The 3.1kHz (300-3400Hz) bandwidth chosen for telephone is the minimum needed for the listener to understand what is being said (but with very limited ability to identify the speaker). 8kHz bandwidth allows better understanding and the ability to identify the speaker.

[limefc]

Using MP3 CBR on voice is pretty damn dumb. VBR or ABR if you're using LAME.

[Saras]

Most podcasts I have are in 64kbps and a few in 128kbps. I'd say 64kbps is enough for speech alone.

[Pentium100]

Most podcasts I have are in 64kbps and a few in 128kbps. I'd say 64kbps is enough for speech alone.

For speech 32k can be enough, maybe even 16k for mono recordings.

[Slysoft]

Why don't you just use ogg? it sounds way better than mp3 at those low bitrates

[nstgc]

Why don't you just use ogg? it sounds way better than mp3 at those low bitrates

My DAP would convert it to WMA. It ends up as one of those three and has to be converted.

[Slysoft]

in that case it just sounds like you have a case of the "don't have a cowon" syndrome

[rkruger]

You could probably also downsample to a 22050 Hz sampling frequency to save space. That sets the Nyquist frequency at ~11kHz, which is still above the upper bound of the voice range.

[nstgc]

in that case it just sounds like you have a case of the "don't have a cowon" syndrome

I wanted a Cowon at one time, but I'm very pleased with my Zune subscription. I like the ability to download music as I want without breaking any laws.

[edit]

You could probably also downsample to a 22050 Hz sampling frequency to save space. That sets the Nyquist frequency at ~11kHz, which is still above the upper bound of the voice range.

That would be a good idea. I have no idea how to do it, but that would probably work (not sure if my DAP can play it, but I could check).

[Pentium100]

You could probably also downsample to a 22050 Hz sampling frequency to save space. That sets the Nyquist frequency at ~11kHz, which is still above the upper bound of the voice range.

16kHz sampling rate would set the Nyquist frequency at 8kHz, which would still be enough. However, since 16kHz is not a divider of 44.1kHz, check to see if your device supports it.