There is mainly two most inportant quanization techniques
a) Wave Codecs
b) Source Codecs

Wave Codecs
G.711 falls in wave codec category and takes bandwidth of 64Kbps approx by using Pulse Code Modulation technique(PCM). It is the same as described by Nyquist. G.726 uses 32,24 or 16 Kbps and uses Adapative Differential Pulse Code Modulation (ADPCM). When G.726 uses 32 Kbps during that time it takes 4 bits for error control and slips the reoccurence bits, for 24 Kbps it uses 3 bit and for 16 Kbps it uses 2 bits. Definately during the less error control bits increases the error rate because you can skips those bits also which need to be used. This is aka quantization error.

Source Codecs
Sorce codecs are designed to work with human voice. Like cisco G.729 takes 8Kbps of bandwidth to deliver voice. Now the question comes in mind why it takes so less? Actually it is having in built codebook in the database and on the basics of frequeny it matches the code and place the code in a packet. Lets assume if you want to say CISCO, it will go like “ccceessccoo” and it is having built in binary for the words. Thats why it eats less bandwidth rather than other protocol.

As move in the less bandwidth protocols the delay need to be considered because it always increases. So one can say bandwidth is inversely proportional to delay.