annotate - automatic speech annotation by transcription alignment


annotate (-i item) (-2) (-d dictionary) (-p prondict) (-t transcription) (-e) file


annotate performs a non-linear time alignment between a coefficient item in the specified file and a coefficient item 'built' from a supplied transcription using a coefficient dictionary. The resulting time alignment is used to construct an annotation item in the file, aligned to the original signal.

The dictionary is simply an SFS file with a number of coefficient items, one per transcription symbol. To identify the coefficient items they should have a 'label=<symbol>' history component. The program andict generates a file in the appropriate form. If no dictionary is specified on the command line the program looks for a file 'dict' in the current directory.

Options and their meanings are:

-I Identify program and version.

-i item Select input item numbers.

-d dictionary Use the specified file as the symbol dictionary rather than 'dict'.

-t transcription Specify the transcription to be aligned. Use quotes to enclose the transcription and a space between each symbol. If no transcription is specified the 'token' field in the main header is used instead.

-p prondict Specify that the transcriptions are given in orthography, and that the correct transcriptions can be found by searching this text file of transcriptions. The file has one transcription per line with the orthography, a TAB, and the transcription.

-2 Apply slope constraints of 2:1 on the DP match. This limits compression or expansion during matching.

-e Remove the mean from the coefficient vectors before matching, adding the mean as an additional scalar value to the vector. This is useful if filterbank energies are being used as coefficients.



11.xx Coefficient item.


11.xx Coefficient item matching dictionary.


5 Time aligned annotation item in data file.


2.2 - Mark Huckvale


vcalign andict
Fri Jul 09 14:54:41 2004