The design of an unusually simple fabric-based touch andpressure sensor is introduced. An analysis of the raw sensor data is shown to have significant non-linearities and non-uniform noise. Using support vector machine learning and a state-dependent adaptive filter it is demonstratedt that these problems can be overcome. The method is evaluated quantitatively using a statistical estimate of the instantaneous rate of information transfer. The SVMregression alone is shown to improve the gesture signalinformation rate by up to 20% with zero added latency, andin combination with filtering by 40% subject to a constant latency bound of 10 milliseconds.