Natsume is a toolkit for Japanese text frontend processing. It's based on the open source project OpenJTalk and its python wrapper pyopenjalk.
Natsume mainly focuses on grapheme-to-phoneme (g2p) conversion and accent annotation. Since the NJD features in OpenJTalk are already enough for determining the boundary and accent nucleus of an accent phrase, Natsume doesn't use full context labels for simplicity and accuracy.
Also, Natsume provides some other functions, including one converting Japanese new fonts to old fonts.
- C/C++ compilers
- cmake
- cython
-
Mac OSX
-
Linux
pip install natsumeIf you're having problems installing, please refer to common problems
from natsume import Natsume
frontend = Natsume()
text = "天気がいいから、散歩しましょう。"Romaji
Convert text to Romaji.
phonemes = frontend.g2p(text, phoneme_mode="romaji", token_mode="phrase")
print(" ".join(phonemes))teNkiga iikara , saNpo shimasho: .
IPA
Convert text to IPA (broad transcription).
phonemes = frontend.g2p(text, phoneme_mode="ipa", token_mode="phrase")
print(" ".join(phonemes))teNkiga iikaɾa , saNpo ɕimaɕo: .
With Accent
Convert text to phonemes with accent.
phonemes = frontend.g2p(text, phoneme_mode="romaji", token_mode="phrase", with_accent=True)
print(" ".join(phonemes))teꜜNkiga iꜜikara , saꜛNpo shiꜛmashoꜜ: .
Get intermediate MeCab features.
from natsume import Natsume
frontend = Natsume()
text = "人間は「食べて寝て」を繰り返すと牛になる。"
mecab_features = frontend.text2mecab(text)
for mecab_feature in mecab_features:
surface = mecab_feature["surface"]
feature_string = ",".join(list(mecab_feature.values())[1:])
print("{}\t{}".format(surface, feature_string))人間 名詞,一般,*,*,*,*,人間,ニンゲン,ニンゲン,0/4,C2
は 助詞,係助詞,*,*,*,*,は,ハ,ワ,0/1,名詞%F1/動詞%F2@0/形容詞%F2@0
「 記号,括弧開,*,*,*,*,「,「,「,*/*,*
食べ 動詞,自立,*,*,一段,連用形,食べる,タベ,タベ,2/2,*
て 助詞,接続助詞,*,*,*,*,て,テ,テ,0/1,動詞%F1/形容詞%F1/名詞%F5
寝 動詞,自立,*,*,一段,連用形,寝る,ネ,ネ,1/1,*
て 助詞,接続助詞,*,*,*,*,て,テ,テ,0/1,動詞%F1/形容詞%F1/名詞%F5
」 記号,括弧閉,*,*,*,*,」,」,」,*/*,*
を 助詞,格助詞,一般,*,*,*,を,ヲ,ヲ,0/1,動詞%F5/名詞%F1
繰り返す 動詞,自立,*,*,五段・サ行,基本形,繰り返す,クリカエス,クリカエス,3/5,*
と 助詞,接続助詞,*,*,*,*,と,ト,ト,0/1,形容詞%F1/動詞%F1
牛 名詞,一般,*,*,*,*,牛,ウシ,ウシ,0/2,C3
に 助詞,格助詞,一般,*,*,*,に,ニ,ニ,0/1,動詞%F5/形容詞%F1/名詞%F1
なる 動詞,自立,*,*,五段・ラ行,基本形,なる,ナル,ナル,1/2,*
。 記号,句点,*,*,*,*,。,。,。,*/*,*
Convert new fonts to old fonts and vice versa.
Please note that 「弁」 has several old fonts. Now, for simplicity, Natsume converts it to 「辯」.
from natsume import Natsume
frontend = Natsume()
new = "桜、桜、うたかたに。"
old = frontend.convert_fonts(new, reverse=False)
print(old)櫻、櫻、うたかたに。
- Natsume: GPL license (LICENSE)
- pyopenjtalk: MIT license (LICENSE.md)
- OpenJTalk: Modified BSD license (COPYING)