Allophone Symbol

Allophone

Format of the string representation of an allophone:

C{L+R}B

where

C is the central phoneme
L is the left phoneme context (or the boundary symbol #)
R is the right phoneme context (or the boundary symbol #)
B specifies allophones at word boundaries: @i for the first phoneme (initial), @f for the last phoneme (final)

Examples

n{a+s}
s{n+#}
s{n+#}@f

Note/Bug: If the phoneme or allophone string contains one of the following symbols:

+ = allophone-neighbor delimiter
- = legacy delimiter
{ = opening allophone-context delimiter
} = closing allophone-context delimiter
# = boundary symbol
. = allophone-state delimiter
todo: there might be more symbols here

then a dumped lookup table probably cannot be read in again. See also: Create State Tying Lookup Table and Bliss Lexicon.

This is especially the case if the RWTH-ASR system is used for OCR tasks, where e.g. punctuations are used in the orthography and resulting phonetic transcriptions.

b-P0{a-P0+c-P0} can be dumped but not read
+{+++} can be dumped but not read
#{#+#} can be dumped and read but is probably not that what you wanted, i.e. a model of the written sharp symbol.

Allophone State

Format of the string respresentation of an allophone state:

A.S

where

A is the Allophone string respresentation
S is the state index (starting from 0)

Examples:

n{a+s}.0
n{a+s}.1
n{a+s}.2

Note: avoid using the state-delimiter symbol as phoneme symbol.

.{.+.}@f.13 representing three dots in a row and modeled with a 14 state HMM can be dumped but probably not read