Who do we know that has expertise here:

''Voice recognition is increasingly noise-robust and accurate on city streets, in cars and in living rooms. Sophisticated signal processing distinguishes between sounds that matter and those that do not, and voice biometrics helps determine who is speaking. Multi-microphone arrays can dynamically steer “listening beams,” which, with the aid of video cameras, can track the location of the user. We are thus moving towards mobile devices that are more aware of the user and his or her context, and are thus more discriminating.''

Machine learning via input and past relational matrixes = Templates.

''The sophistication of NLU – inferring intent from a user’s input and acting on it – is also increasing. Today’s state-of-the-art blends powerful linguistic frameworks (called symbolic processing), explicit representation of knowledge (ontologies) and machine learning that take advantage of big data to populate the frameworks with observed examples and patterns. This complementarity is important: past symbolic approaches proved to be fragile, and pure data-driven NLU typically only achieved shallow understanding''

e