find the k nearest neighbour templates using DTW for a given context dependent phone graph More...

Detailed Description

find the k nearest neighbour templates using DTW for a given context dependent phone graph

spr_knn_dtw [-LOO](flag: leaving-one-out) <-ref reference data> <-seg label file>
    [-wrd word prefix] <-tst test data> [-mvg gaussian set] [-mvgf factor](-0.5)
    [-ndc cost](10.0) <-lat the lattice file> <-o knn file> [-k k](15) [-k1 k1](1)
    [-threads #threads](0) [-ctx #frames](0)

Parameters

	-LOOflag	leaving-one-out Do not compare with templates originating from the same speaker as the current test speaker (leaving-one-out); assumes that, except for data sharpening, the test equals the reference data.
	-ref<em>reference	data The reference data.
	-seg<em>label	file The segmentation file, list the templates and assigns labels to them.
	-wrd<em>word	prefix Convert the words to templates as well, prefix with the given string. The word template numbering (template index) starts with (#phone_templates) and counts up from there.
	-tst<em>test	data The test data.
	-mvg<em>gaussian	set The gaussian set use to calculate the sensitivity (variance) on the test data.
	-mvgf<em>factor</em><a	name="spr_knn_dtw.mvgf" class="el"> Factor used to convert the distance to a probability for the sensitivity data. Use -0.5/F for correlated data where F>=1.0 indicates the degree of correlation.
	-ndc<em>cost</em><a	name="spr_knn_dtw.ndc" class="el"> Cost for non-diagonal transitions.
	-lat<em>the	lattice file The (pseudo) lattice file listing all segments in the test data which must be handled.
	-o<em>knn	file The output file containing the knn nearest neighbours.
	-k<em>k</em><a	name="spr_knn_dtw.k" class="el"> Number of neighbours to calculate.
	-k1<em>k1</em><a	name="spr_knn_dtw.k1" class="el"> Number of neighbours to calculate when DTW fails (extreme warping).
-threads<em>\	threads</em><a	name="spr_knn_dtw.threads" class="el"> Number of threads used to compute the knn's.
-ctx<em>\	frames</em><a	name="spr_knn_dtw.ctx" class="el"> Context extension (symmetric).

Find the k nearest neighbour templates using DTW for a given context dependent phone graph.

Author: Kris Demuynck

Date: 23/07/2010

Revision History:

23/07/2010 - K.D.: creation
20/08/2010 - K.D.: add mahanalobis distance