At the core of this library is text normalization and word stressing processor from LIEPA speach synthesizer. The native code related to text processing was cut out of the synthesizer library code and wrapped in Python.
The library takes text in Lithuanian and does following:
- Normalizes it. Converts numbers to word reprezentations (e.g. "1" > "vienas").
- Splits text into phrases/sentences.
- Splits phrases into words
- Identifies word syllables
- Identifies possible grammar forms of the word, and identifies stressed letter and stress type according the grammar form
- Chooses one rule
- Returns either structured results or collapsed
Library supports following environments:
- Python: 2.7, 3.*
- OS: Linux, Windows
- Architecture: 32bit, 64bit
pip install phonology_engine
Conversion from numbers to word representation.
from phonology_engine import PhonologyEngine
pe = PhonologyEngine()
res = pe.normalize_and_collapse('31 kačiukas perbėgo kelią.')
print(res)
Would result in
TRISDEŠIMT VIENAS KAČIUKAS PERBĖGO KELIĄ.
Determining word stresses.
from phonology_engine import PhonologyEngine
pe = PhonologyEngine()
res = pe.process_and_collapse('31 kačiukas perbėgo kelią.', 'utf8_stressed_word')
print(res)
Would result in
TRÌSDEŠIMT VÍENAS KAČIÙKAS PÉRBĖGO KẼLIĄ.
Determining word stresses, syllables, grammar form from word.
from phonology_engine import PhonologyEngine
from pprint import pprint
pe = PhonologyEngine()
res = pe.process(u'31 kačiukas perbėgo kelią.')
for word_details, phrase, normalized_phrase, letter_map in res:
for word_detail in word_details:
pprint (word_detail)
Would result in
...
{'ascii_stressed_word': 'TRI`SDEŠIMT VI^ENAS',
'normalized': True,
'number_stressed_word': 'TRI0SDEŠIMT VI1ENAS',
'span_normalized': (0, 17),
'span_source': (0, 2),
'stress_options': {'decoded_options': [{'rule': 'Nekaitomas žodis'}],
'options': [(2, 0, 1, 1688)],
'selected_index': 0},
'syllables': [0, 3, 6],
'utf8_stressed_word': 'TRÌSDEŠIMT VÍENAS',
'word': 'TRISDEŠIMT VIENAS',
'word_with_all_numeric_stresses': 'TRI0SDEŠIMT VI1ENAS',
'word_with_only_multiple_numeric_stresses': 'TRISDEŠIMT VIENAS',
'word_with_syllables': 'TRI-SDE-ŠIMT VIE-NAS'}
{'ascii_stressed_word': 'kačiu`kas',
'normalized': True,
'number_stressed_word': 'kačiu0kas',
'span_normalized': (18, 26),
'span_source': (3, 11),
'stress_options': {'decoded_options': [{'grammatical_case': 'Vardininkas',
'number': 'vienaskaita',
'rule': 'Linksnis ir kamieno tipas',
'stem_type': 0,
'stress_type': 0,
'stressed_letter_index': 4}],
'options': [(4, 0, 2, 0)],
'selected_index': 0},
'syllables': [0, 2, 5],
'utf8_stressed_word': 'kačiùkas',
'word': 'kačiukas',
'word_with_all_numeric_stresses': 'kačiu0kas',
'word_with_only_multiple_numeric_stresses': 'kačiukas',
'word_with_syllables': 'ka-čiu-kas'}
{'ascii_stressed_word': 'pe^rbėgo',
'normalized': False,
'number_stressed_word': 'pe1rbėgo',
'span_normalized': (27, 34),
'span_source': (12, 19),
'stress_options': {'decoded_options': [{'rule': 'Veiksmazodžių kamienas ir '
'galune (taisytina)'}],
'options': [(1, 1, 0, 465)],
'selected_index': 0},
'syllables': [0, 3, 5],
'utf8_stressed_word': 'pérbėgo',
'word': 'perbėgo',
'word_with_all_numeric_stresses': 'pe1rbėgo',
'word_with_only_multiple_numeric_stresses': 'perbėgo',
'word_with_syllables': 'per-bė-go'}
{'ascii_stressed_word': 'ke~lią',
'normalized': False,
'number_stressed_word': 'ke2lią',
'span_normalized': (35, 40),
'span_source': (20, 25),
'stress_options': {'decoded_options': [{'grammatical_case': 'Galininkas',
'number': 'vienaskaita',
'rule': 'Linksnis ir kamieno tipas',
'stem_type': 2,
'stress_type': 2,
'stressed_letter_index': 1}],
'options': [(1, 2, 2, 515)],
'selected_index': 0},
'syllables': [0, 2],
'utf8_stressed_word': 'kẽlią',
'word': 'kelią',
'word_with_all_numeric_stresses': 'ke2lią',
'word_with_only_multiple_numeric_stresses': 'kelią',
'word_with_syllables': 'ke-lią'}
- Kirčiavimas internetu - Online dictionarry with word stresses and grammar annotation, has a GitHub repo. It is likely based on VDU dictionary.