The signal (amplified or unamplified speech) to noise (all other sound) ratios at the listener position should be better than 25 dB to optimise intelligibility.
In some environments where high levels of background noise are present, this is still a good target to aim for but may not be practical (eg. 75dB background noise plus 25dB of signal would be 100dB overall, which is too loud to be comfortable for the users of the space). In these situations other alternatives such as closed-ear hearing augmentation could be considered.
It is worth noting that SNR is a generic ratio, applied to a wide range of acoustic and electrical system measurements. Therefore care must be taken to design to the correct SNR for each application.