Today's TDM networks represent audio in G.711 format, also known as the mu-Law or A-law codec.  This codec has been around since the 1970's when digital networks started to carry audio as digitised data.  According to sampling theory you need to sample at least twice as fast as the highest frequency that you want to preserve, so 8000 samples/sec was chosen as the sampling rate, allowing frequencies of less than 4kHz to be represented.  Since telephone networks have been engineered for a long time to use only part of the voice spectrum (perhaps 3.5kHz), then this was a suitable engineering compromise for the digital age.

I've heard several people refer to A-law and mu-Law as "uncompressed" audio, but this is not at all true.  Actually, each 8-bit sample is split into two parts, an exponent and a mantissa (if you've ever done any work on representing real numbers in a computer you'll be familiar with this was of working).  So the exponent gives the overall magnitude, and the mantissa gives the fine detail.  It's a logarithmic representation, which happens to work well for the human ear, and usefully it allows us to pack 12 or 13 bits of resolution into only 8 bits. 

Coding and decoding for G.711 is a relatively cheap operation (for a CPU), which must have been a deciding factor back in the early days of computing.  For example, the A-Law decoder looks like this:

  absval = (sample>>4) & 7;       //exponent
  lower  = (sample & 0x0F)<<1;    //lower 4 bits of value

     case 0: amplitude =   lower|0x01;     break;
     case 1: amplitude =   lower|0x21;     break;
     case 2: amplitude =  (lower|0x21)<<1; break;
     case 3: amplitude =  (lower|0x21)<<2; break;
     case 4: amplitude =  (lower|0x21)<<3; break;
     case 5: amplitude =  (lower|0x21)<<4; break;
     case 6: amplitude =  (lower|0x21)<<5; break;
     case 7: amplitude =  (lower|0x21)<<6; break;

So for a handful of shifts, ORs and ANDs you can recover the audio sample into PCM format (i.e. really uncompressed) that is suitable for whatever processing comes next (saving to file, DTMF detection etc).

Now in the VoIP age, people are starting to feel limited by G.711 resolution, and in fact many VoIP systems allow alternative codecs to be used.  Between VoIP and mobile there has been an explosion of development of many different codecs in the last 20 years, using the much more powerful and ubiquitous CPUs that we now have.  Wideband codecs are now increasing in popularity, for example AMR-WB (G.722).  Wideband codecs have a wider sample size, so that more frequencies can be represented, making the audio more akin to broadcast radio in terms of sound quality. There are even technologies like TFO (transcoder free operation) that aim to retro-fit wideband technology to existing TDM networks.  Ultimately we can imagine the humble telephone producing broadcast quality audio, and that will be a huge step forward in usability.