Select frames corresponding to units or states, or to all data.
- Parameters
-
-ssp<em>signal | processing script
The signal processing script used to preprocess the input data. |
-c<em>corfname</em><a | name="spr_sel_frames.c" class="el">
Input corpus file name. |
-sel<em>sel_string</em><a | name="spr_sel_frames.sel" class="el">
Unit name or state name (unit#nr). All frames corresponding to this string are copied to the output file. |
-u<em>unitfilename</em><a | name="spr_sel_frames.u" class="el">
Unitfile name with all units for which frames should be selected (.arcd or .cd format). If unit of state based selection is asked for, or option -u should be given, or (with lower priority), option -sel. |
-o<em>ofname</em><a | name="spr_sel_frames.o" class="el">
Output file name. If the selected frames are based on units or states, a second file, with same name but with extention 'siz' will give the length of all parts. |
-suffix<em>file_extension</em><a | name="spr_sel_frames.suffix" class="el">
File name extention for an entry in the corpus file. |
-margin<em>margin</em><a | name="spr_sel_frames.margin" class="el">
Frames are only copied starting from (Firstframe + margin) to (Lastframe - margin). |
-nsel<em>NrSelected</em><a | name="spr_sel_frames.nsel" class="el">
Number of frames to select per state/unit/corpus. Default all. |
-nfr_wanted<em>nfr_wanted_filename</em><a | name="spr_sel_frames.nfr_wanted" class="el">
Unitfile-like file with on each line a unit or state describing string and the number of frames wanted for that unit or state. Default is given with option -nsel. The number of lines can be indicated with the NLINES key. |
-reduc<em>reduction</em><a | name="spr_sel_frames.reduc" class="el">
Reduction factor for the files in the corpus. Only every 'cor_reduc' different files, a file will be processed. Float values can be used, avoiding that in some cases, only files of one sort would be chosen. |
-base<em>BaseFlag</em><a | name="spr_sel_frames.base" class="el">
Says if selection should be unit based (U), state based (S), or if frames should be selected from all data (C). |
-obs<em>ObsDir</em><a | name="spr_sel_frames.obs" class="el">
Directory in which the files, listed in the corpus file, can be found. |
Select frames corresponding to units or states, or to all data. This program extracts frames out of the data files listed in a corpus or segmentation file.
If the selection is corpus based (-base C), a number of frames from all data will be selected (option -nsel). If the selection is unit or state based (-base U or S), for each unit/state asked for, a number of frames (option -nsel and nfr_wanted) corresponding to that unit/state will be selected. Choose one unit/state with option -sel, or more by giving a unit file.
All frames are written in one file. If frames corresponding to units or states are asked for, a second file gives begin and end frame for each unit or state. If there are a lot of different files in the corpus file, one can ask to select e.g. 5 frames for each fifth different file in the corpus, instead of 1 frame for each file (option -reduc). It is possible to give a segment boundary margin so that frames within this margin from the boundary of a segment are not extracted.
- Author
- Jacques Duchateau
- Date
- 14/06/1994
- Revision History:
- 09/94 - JD
- changes for later use with preprocessing subroutines
- 04/95 - KD
- on-line preprocessing added.
- 01/96 - JD
- adapted for new hmm structures.