I'm sure this was covered somewhere, but I searched and didn't come up with anything.

My question is what Bit Rate/sample rate to encode a male's voice in. Basically I would like to compress a lecture, and since space is at a premium I would like to use a relatively low bit rate (I guess somewhere between 24kbps-80kbps) I also would like to use CBR. Of course I don't want to go so low that it sounds really bad, so the question is at what Bit Rate does a male's voice sound OK, I'm not looking for transparency, just normal-listenable quality.

Also what sample rate does it make sense to encode in? I understand that in general higher sampling is better but at such relatively low Bit Rate if I use say 44100 won't that leave only a tiny amount of data for each sample, would it therefore be better to use a lower sampling rate say 22050, 24000 or 32000? that way the each sample size will be bigger, (or does it not work like that at all, I hope this make some sense).

for human voice 22050 is good enough

Human vocal's pitch will not exceed 11025 Hz?

this is what i understand about sample rate, a 44100 Hz sample rate can store a maximum frequency of 22050 Hz....as for 22050, it can store up to 11025 Hz!

Anyway, why we need 2 samples to represent 1 frequency?? confusing
the frequency stored is always half of the sample rate, why???

I'm currently involved in a similar project: I'm converting a series of lecture tapes to digital format.

The settings I decided on are:
1 - Record the source to a 22050-Hz, 16-bit mono WAV file;
2 - Convert the WAV to FLAC for archiving;
3 - Convert the FLAC to MP3 using LAME 396.1 set to -V8 --vbr-new.
or, alternatively
3 - Convert the FLAC file to OggVorbis (v1.1) set to -q1

Using these settings, a 45-minute lecture results in a ~70M FLAC file, a ~14M MP3 (at ~40kbps), and a ~11M OggVorbis file (at ~30kbps). I also found that if I use noise reduction to remove some of the tape hiss, I can further reduce the file size by 10-20%.

These were the best settings I could find without degrading the sound quality.

Hope this help a bit.



~esa

If you're willing to bend on the CBR requirement (does your player have trouble seeking VBR files?), you could try LAME with the voice optimized preset "--preset voice" (or the equivalent "--preset 56 -m m") which is 32kHz mono, 56kbps avg... It sounds great to me, though you might be happy with even lower bitrate (and space saving) options.

similar discussion with more info / options:
http://www.hydrogenaudio.org/forums/index....&showtopic=9449
http://www.hydrogenaudio.org/forums/index....showtopic=17526
http://www.hydrogenaudio.org/forums/index....showtopic=20043
http://www.hydrogenaudio.org/forums/index....&showtopic=3270
http://www.hydrogenaudio.org/forums/index....showtopic=23282
http://www.hydrogenaudio.org/forums/index....showtopic=28821

I've yet to come accros any lossy encoder that doesn't degrade quality at that bitrate

Quite surprising for me such long practiced HA member doesn't understand Nyquist theory.
Maybe that will explain something.

Quite right...
Allow me to amend my statement:

"These were the best settings I could find without degrading the sound quality too terribly."

The Wikipedia article is very informative:
http://en.wikipedia.org/wiki/Shannon-Nyqui...ampling_theorem
the Mathworld page is very minimal, but includes the essentials:
http://mathworld.wolfram.com/SamplingTheorem.html