perm filename FILES.DOC[11,ALS] blob
sn#065558 filedate 1973-10-05 generic text, type T, neo UTF8
00010 Useful Files on 11,ALS 10/5/73
00020
00030 ADDNAM - Allowa one to add names to the list of input pameters.
00040
00050 ADJUST -Uses data produced by file NORMAL to adjust the input data
00060 for the specified files.
00070
00080 BLOCKS- Not an operating program but a useful required item for a
00090 program. BLOCKS establishes the necessary arrays to hold the
00100 information in file TABHED.DAT and contains a callable routine HEADIN
00110 which loads these arrays.
00120
00130 CLUST - Produces cluster diagrams for up to 4 different Phonettes or
00140 Features plotted for any 2 specified input parameters. Charts are
00150 produced showing the distribution of data for each Phonette or
00160 Feature separately, then there is a simngle plot showing regions
00170 dominated by the different Phonettes, and then confusion plots and
00180 finally plots showing the amount of submerged data. A list must be
00190 supplied of the files that are to be used as inputs.
00200
00210 CONVER - Converts old style headers into the newer form. This program
00220 will have outlived its usefulness as soon as the new headers have
00230 been stabilized.
00240
00250 EXTEND - A program to add input parameter information to the headers.
00260 At the present time this program is fixed to add frication and pitch
00270 period information but the core of the program can be easily modified
00280 for any other type of information.
00290
00300 FLOP4 - This program computes the number of reversals in the 4th
00310 difference obtained for the specified input acoustic wave for each
00320 6.4 millisecond portion. This figure is used as a measure of
00330 Frication. Its usefulness has not yet been demonstrated altho it does
00340 clearly differentiate between utterances having very little high
00350 frequencies as compared with those that do.
00360
00370 HEAD - Prepares a document showing the data in the header for any
00380 file. The file for this document may be modified using any
00390 conventional Editor in order to correct the phonettes, their starting
00400 locations and lengths and then this corrected file may be used as a
00410 source file by program UPDATE to correct the header of an input file.
00420
00430 HISSEG -A histograph producing program.
00440
00450 LEARN - The learning program which accumulated data derived from
00460 annotated input files. At the present time LEARN requiires
00470 preprocessed input files in which 24 different parameters have been
00480 tabulated for each 6.4 millisecond portion of the original acoustic
00490 input. This file must also contain a header as prepared by UPDATE
00500 with the so-called HINTS, these being the specification of the name,
00510 starting position and length of clearly recognizable portions of the
00520 input, again measured in units of 6.4 milliseconds.
00530
00540 LEVEL2 - The start ol a second level program to take account of
00550 context to modify the segment probabilities that are computed by a
00560 SEG type program. This version compiles but it has not yet been tested.
00570
00580 LRN.DAT- This file contains the currently accumulated learning counts
00590 during the operation of LEARN. Overflow counts are kept in file
00600 RES.DAT. and the data used by the recognition programs is stored in
00610 file USE.DAT. For further details see USE.DAT below.
00620
00630 MAKE - Provides facilities for creating and modifying SIGNATURE TABLE
00640 header information. The program displays the existing statis of the
00650 tables and prompts with questions so that the nacessary information
00660 can be supplied in the right form and order. The program creates the
00670 required machine-readable header file TABHED.DAT and it then creates
00680 a man-readable document file TABLES.DOC which reports on the final
00690 state of the header information.
00700
00710 MAP.PHN - Miller's map of Phonettes illustrated by examples and
00720 showing the associated significant features.
00730
00740 NAMEIN - Provides facilities to add names to the list of input
00750 parameters. See ADDNAM above which does nearly the same thing.
00760
00770 NORMAL - Produces a table that can be used to normalize the input
00780 files to preserve the maximum amount of resolution when the allowable
00790 range of values is reduced by using less than the full 6 bits allowed
00800 by the inputs.
00810
00820 PIT -
00830 PITCH -
00840 PPITCH- Various pitch and glottal pulse detection programs, all in a
00850 very preliminary form, They will undoubtedly be superseded by Niel's
00860 program if and when this is working.
00870
00880 PLOT - Lists header information from .T0X files on [11,ALS] and shows
00890 the corresponding acoustic data from files on [CMP,JH]. The program
00900 will select examples for any specified phonette or it cam be used to
00910 display everything, The method of use and the different options that
00920 are available are described on entry.
00930
00940 RES.DAT - This is the overflow file refered to in LRN.DAT above and
00950 described further in USE.DAT below.
00960
00970 SEE -Displays a summary of the information contained in the SIGNATURE
00980 TABLES.
00990
01000 SEER - Similar to SEE but producing a hard copy.
01010
01020 SEG -A simplified USE program written for use in obtaining output
01030 data for the Segmentation Workshop. The data presented at Pittsburg
01040 was obtained by this program. This program requires preprocessed files
01050 which are currently filed under the nemes SEG1.T0X, SEG2.T0X etc. and
01060 which contain 24 6-bit parameters for each 6.4 millisecond portion of
01070 the original acoustic input data.
01080
01090 SEG1.T0X - The first of a group of files (currently 31) as described
01100 more fully under SEG above.
01110
01120 SIG.MAC - The assembly language portion of the signature-table
01130 operating system, which does the inner loop for both the learning
01140 routine and the use routine. SIG is called from these routines with a
01150 parameter to specify the type of operation to be performed.
01160
01170 STUDY4 - A study program for investigating simple parameter extraction
01180 routines as applied to acoustic data and tabulating the results. It is
01190 currently set up to report on a FLOP4 routine which counts the number
01200 of reversals in direction of the fourth difference over each 6.4
01210 millisecond interval of the acoustic wave.
01220
01230 TABHED.DAT- Contains the so-called TABLE-HEADER for a set of
01240 SIGNATURE TABLES. It lists input information and has all of the
01250 necessary interconnection information for the tables. TABHED.DAT is
01260 created be the program called MAKE which is separately described.
01270
01280 UPDATE- Uses a file document made by program HEAD, corrected or not
01290 as noted above to update the information in the header of a specified
01300 file.
01310
01320 USE.DAT - Machine readable data produced by program LEARN and used by
01330 program SEG. This file contains only the derived probability figures.
01340 The original data that is obtained during training sessions is contained
01350 in two additional files RES.DAT AND LRN.DAT. The file LRN.DAT has
01360 room for counts up to 511 while file RES.DAT counts only in units of
01370 32. The top 4 bit positions in the counts accumulated in LRN.DAT are
01380 periodically added to the counts in RES.DAT and then reset to zero to
01390 allow for further accumulations. By this dodge it is possible to allow
01400 counts up to 2↑14 while limiting the size of the tables that must be
01410 maintained in the learning program.
01420