Methods and systems for testing speech recognition systems are disclosed in which the speech recognition device to be tested is directly monitored in accordance with a text-to-speech device. The device includes a manifold having an inlet for connection to the case drain. Information about the detected communication attempt is collected. The dielectric material may form a layer in the trenches located adjacent to each other of substantially equivalent height on both sides of the unetched portion of the substrate separating the two trenches.