We introduce an elegant and novel design for a speech detector which estimates the probability of the presence of speech in each time-frequency bin, as well as in each frame. The proposed system uses discriminative estimators based on logistic regression, and incorporates spectral and temporal correlations in the same framework. The detector is flexible enough to be configured in a single level or a “stacked” bi-level architecture depending on the needs of the application. An important part of the proposed design is the use of a new set of features: the normalized logarithm of the estimated posterior signal-to-noise ratio. These can be easily and automatically generated by tracking the noise spectrum online. We present results on the AURORA database to demonstrate that the overall design is simple, flexible and effective.