Grapheme to phoneme rules for Estonian
Implements a naive rule-based G2P converter for Estonian.
Note:
- Compound words should have their compound constituents separated with and underscore in input (e.g. 'hamba_kivi')
Compile
ant
Test
ant test
Run
$ echo "kesk_pank" | ./run.sh
kesk_pank k e s k p a n kk
$
The script reads words from stdin and writes words with their pronunciations back to stdout. Usually it's used to process a long list of words at a time:
$ ./run.sh < example/sample.vocab
kana k a n a
park p a r kk
näinud n ae i n u t
näinud(2) n ae i n t
kontsa k o n t s a
kesk_pank k e s k p a n kk
OECD o o e t s e e t e e
ETV24 e t e e v e e k a k s k ue m m e n t n e l i
ETV24-le e t e e v e e k a h e k ue m n e n e l j a l e
NATO n a tt o
ABC-pood a a p e e t s e e p o o t
René r e n e
Poincaré p o i n kk a r e
$
The script can be also executed from any other directory:
~/tools/et-g2p/run.sh < vocab.txt
The tool supports user-defined transliteration dictionaries. This allows to define pronunciations for words that are pronounced differently from the Estonian rules and which are not defined in the built-in exception list.
Each line in the dictionary contains the original word and its transliteration (i.e., its probable ortographic form as if it was an Estonian word). Multiple transliterations can be given, seperated by commas. For example:
Jules žül
Henri henri, anrii
Poincaré puangarree
Use the -dict option to set the user dictionary:
echo "Henri" | ./run.sh -dict tmp.dict
Henri a n r i i
Henri(2) h e n r i
Bugs
Pronunciation of foreign names is based on a short list of exceptions. As a result, for most
English and French names, pronunciation is generated according to Estonian rules, which is
of course wrong (see Poincaré above).
Use as a library
See src/java_test for samples.