Construction of acoustic models given a decision tree. More...

Detailed Description

Construction of acoustic models given a decision tree.

spr_tree2model [-global](flag: ) <-hi HMMFileName> <-ho HMMFileName> [-ui UnitFileName]
    <-uo UnitFileName> [-gi MVGFileName] <-go MVGFileName> [-seli SelFileName]
    <-selo SelFileName> <-ci CIphones> <-cdi CDphones> <-cdo CDphones> <-q Questions>
    [-counts State_counts] <-tree Trees> [-context_len Context_length](1)
    [-min_count Min_mixture_count](1) [-max_overlap Max_overlap](0.0)

Parameters

-globalflag	The tree is a global tree, which means that phones that have no example at all should not be purged.
-hi<em>HMMFileName</em><a	name="spr_tree2model.hi" class="el"> Input HMM with the distributions of the non-tied states.
-ho<em>HMMFileName</em><a	name="spr_tree2model.ho" class="el"> Output HMM with the tied states.
-ui<em>UnitFileName</em><a	name="spr_tree2model.ui" class="el"> Input Unit file name (.arcd or .cd format).
-uo<em>UnitFileName</em><a	name="spr_tree2model.uo" class="el"> Output Unit file name (.arcd format).
-gi<em>MVGFileName</em><a	name="spr_tree2model.gi" class="el"> MVG file name.
-go<em>MVGFileName</em><a	name="spr_tree2model.go" class="el"> Output MVG file name.
-seli<em>SelFileName</em><a	name="spr_tree2model.seli" class="el"> Input SELECT file name.
-selo<em>SelFileName</em><a	name="spr_tree2model.selo" class="el"> Output SELECT file name.
-ci<em>CIphones</em><a	name="spr_tree2model.ci" class="el"> File with context independent phones.
-cdi<em>CDphones</em><a	name="spr_tree2model.cdi" class="el"> Input File with context dependent phones.
-cdo<em>CDphones</em><a	name="spr_tree2model.cdo" class="el"> Output File with context dependent phones.
-q<em>Questions</em><a	name="spr_tree2model.q" class="el"> File with the phonetic questions. The format is described in the manual for routine read_questions().
-counts<em>State_counts</em><a	name="spr_tree2model.counts" class="el"> File with the occupancies, the number of frames assigned to each state in the input HMM. In later package versions, these numbers are stored in the HMM file itself (in this case this file is not necessary).
-tree<em>Trees</em><a	name="spr_tree2model.tree" class="el"> File to read the trees from.
-context_len<em>Context_length</em><a	name="spr_tree2model.context_len" class="el"> Context length (1 for triphones, 2 for quinphones).
-min_count<em>Min_mixture_count</em><a	name="spr_tree2model.min_count" class="el"> Minimal mixture count for both subtrees, not implemented: use the option -min_count in programme tree_create.
-max_overlap<em>Max_overlap</em><a	name="spr_tree2model.max_overlap" class="el"> For tree pruning: gives the maximal overlap between the subtrees (for the MIXTURE criterion) or the maximal log likelihood decrease (for the DISC and SINGLE criteria) that may occur through the splitting of a node in the tree. If the overlap is outside the range [0 max_overlap] with max_overlap <= 1.0 or the log likelihood decrease outside the range [-inf max_overlap] with max_overlap <= 0.0, no further splitting is done.

Construction of acoustic models given a decision tree. Based on a phonetic decision tree (generated with the tree_create programme) and on the statistics for the non-tied states, new tied state HMMs (with the corresponding model context descriptions) are constructed and written to files.

Author: Jacques Duchateau

Date: January 1997

Revision History:: JD, March 1999: added to package HMM version 70