We present a biologically motivated method for assessing the intelligibility of speech recorded or transmitted under various types of distortions. The method employs an auditory model to analyze the effects of noise, reverberations, and other distortions on the joint spectro-temporal modulations present in speech, and on the ability of a channel to transmit these modulations. The effects are summarized by a spectro-temporal modulation index (STMI). The index is validated by comparing its predictions to those of the classical STI and to error rates reported by human subjects listening to speech contaminated with combined noise and reverberation. We further demonstrate that the STMI can handle difficult and nonlinear distortions such as phase-jitter and shifts, to which the STI is not sensitive.