Hi,

as long your network does not drop a single packet, you will be fine with G.729.

However it does not tolerate packet loss, hence the above caveat.

I'm not aware of any issue based on the type of voice carried, it works equally well over a broad range of frequencies and utteration.

G.729 is an amazing codec. Technically referred to as CS-ACELP, or conjugate-structured algebraic code excited linear prediction, this codec actually achieves its incredible fidelity to bandwidth formula by not compressing or sending actual audio at all in its payload. This codec actually sends numeric values that correlate to a specific and pre-populated library or codebook of line-spectrum pairs (LPCs) that are optimized for human voice.

http://www.data-compression.com/speech.shtml

The fidelity of a wide range of human voices (male or female) and even baby sounds or screams or cat meows or dog barks are transmitted over this codec very effectively.

There is essentially no error-correction as a part of the codec so the assumption is that greater than 99% of the encoded packets will arrive at the other end within the 50 to 150 ms window of the playout buffer and echo canceler. Even a 1% error across the transmission medium can cause significant voice depredation with G.729 that may not be as noticeable as an actual audio compression codec like G.711.

This is normally not an issue with any current terrestrial technologies (i.e. everything but satellite links) and therefore is the codec of choice in the vast majority of VoIP designs with low bandwidth links (<1.5Mbit).