From your comment I take it you are looking for speech quality measures. As far as I know, there exist a couple of objective measures, along with subjective procedures, since quality of speech is kind of a vague concept which may not be the same to everyone.
Since your application seems to require automatic evaluation, there are a number of measures you might want to check out:
- ITU-T Rec. P 835 for assessment of listening speech quality (no available reference signal)
- The PESQ for listening speech quality, mentioned by Florian already (when the reference signal is available)
- The PEAQ for assessment of audio quality (when the reference signal is available)
- The E-model für conversational speech quality (network parameters available)
All of these are listed in this book, along with a range of subjective measures.
It seems to me that for your application, the ITU-standard is most reasonable since no clean reference signal is required.
Edit: What also may be of interest of you is the frequency-weighted segmental SNR (fwsegSNR), which incorporates the articulation index (AI). This measure apparently correlates well with subjective measures.
An overview of objective speech quality measures is also given in this Loizou paper available online.