Recognizer to produce structured output vectors.

[hyyrynen]

Previously speech recognition result time values were embedded to the recognition stream with -tags. This commit adds support for producing structured speech recognition output through the swig-interface. The resulting structured stream is a vector of pairs of morphs and start and end time pair values: using namespace std; typedef vector<pair<string,pair<double,double> > > timed_token_stream_type; This structure can be read from python as a list of dictionaries with token, start and endtimes [{'token', 'starttime', 'endtime'}] using the "const timed_token_stream_type &best_timed_hypo_string(bool print_all);"-function

[senarvi]

Looks like this doesn't break anything.

[psmit]

Thanks Lasse!