POLYSEM 1.0 is a program for disambiguating spoken restrictive relatives and quoted vs unquoted expressions. The naive Bayesian classifier (currently implemented) yields a succcess recognition rate of ca 85% (according to my tests with a native speaker of BrE). The Multi-layer Perceptron Classifier (not implemented) yields a success rate of ca 95% (tested on the durations of silent intervals in the senteces produced by the BrE speaker).
The program works if you have all the necessary Python packages installed. It doesn't work without the free software Praat used in speech research.
There is also an auxiliary tool, MLP Optimiser, which is still under development, but it already works for the two datasets (i.e., relative clauses and (un)quoted strings). It was used to search for the best parameters for the Multilayer Perceptron Classifier.