pyopenjtalk-mod

pyopenjtalk-modはpyopenjtalk-plusをベースに改造することを目的とした、pyopenjtalkの派生ライブラリです pyopenjtalkの代替品として性能向上が期待できますが、開発頻度が高いため、pyopenjtalk-plusとの互換性は考えていません

ちなみにpyopenjtalk-plus は、各フォークでの改善を一つのコードベースにまとめ、さらなる改善を加えることを目的とした、pyopenjtalk の派生ライブラリらしいです。

user_dictionary以下の辞書のソースはopenjtalk-user-dictにあります

version log

wikiにあります

Installation

下記コマンドを実行して、ライブラリをインストールできます。

pip install pyopenjtalk-mod

Supported platforms

• Linux
• Mac OSX
• Windows (MSVC) (see this PR)

Build requirements

The python package relies on cython to make python bindings for open_jtalk and hts_engine_API. You must need the following tools to build and install pyopenjtalk:

• C/C++ compilers (to build C/C++ extentions)
• cmake
• cython

Development

開発環境は macOS / Linux 、Python バージョンは 3.11 が前提です。

# submodule ごとリポジトリを clone
git clone --recursive https://github.com/tsukumijima/pyopenjtalk-mod.git

#or
git clone --recursive https://github.com/tsukumijima/pyopenjtalk-mod.git
git submodule update --recursive --init

cd pyopenjtalk-plus

# ライブラリ自身とその依存関係を .venv/ 以下の仮想環境にインストールし、開発環境を構築
pip install taskipy
task install

# コード整形
task lint
task format

# テストの実行
task test

# pyopenjtalk/dictionary/ 以下にある MeCab / OpenJTalk 辞書をビルド
## ビルド成果物は同ディレクトリに *.bin / *.dic として出力される
## ビルド後の辞書データは数百 MB あるバイナリファイルだが、取り回しやすいよう敢えて Git 管理下に含めている
task build-dictionary

# ライブラリの wheel と sdist をビルドし、dist/ に出力
task build

# ビルド成果物をクリーンアップ
task clean

docs/ 以下のドキュメントは、pyopenjtalk 本家のドキュメントを改変なしでそのまま引き継いでいます。

A python wrapper for OpenJTalk.

The package consists of two core components:

• Text processing frontend based on OpenJTalk
• Speech synthesis backend using HTSEngine

Notice

• The package is built with the modified version of OpenJTalk. The modified version provides the same functionality with some improvements (e.g., cmake support) but is technically different from the one from HTS working group.
• The package also uses the modified version of hts_engine_API. The same applies as above.

Before using the pyopenjtalk package, please have a look at the LICENSE for the two software.

Quick demo

Please check the notebook version here (nbviewer).

TTS

In [1]: import pyopenjtalk

In [2]: from scipy.io import wavfile

In [3]: x, sr = pyopenjtalk.tts("おめでとうございます")

In [4]: wavfile.write("test.wav", sr, x.astype(np.int16))

Run text processing frontend only

In [1]: import pyopenjtalk

In [2]: pyopenjtalk.extract_fullcontext("こんにちは")
Out[2]:
['xx^xx-sil+k=o/A:xx+xx+xx/B:xx-xx_xx/C:xx_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:xx_xx#xx_xx@xx_xx|xx_xx/G:5_5%0_xx_xx/H:xx_xx/I:xx-xx@xx+xx&xx-xx|xx+xx/J:1_5/K:1+1-5',
'xx^sil-k+o=N/A:-4+1+5/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'sil^k-o+N=n/A:-4+1+5/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'k^o-N+n=i/A:-3+2+4/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'o^N-n+i=ch/A:-2+3+3/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'N^n-i+ch=i/A:-2+3+3/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'n^i-ch+i=w/A:-1+4+2/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'i^ch-i+w=a/A:-1+4+2/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'ch^i-w+a=sil/A:0+5+1/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'i^w-a+sil=xx/A:0+5+1/B:xx-xx_xx/C:09_xx+xx/D:xx+xx_xx/E:xx_xx!xx_xx-xx/F:5_5#0_xx@1_1|1_5/G:xx_xx%xx_xx_xx/H:xx_xx/I:1-5@1+1&1-1|1+5/J:xx_xx/K:1+1-5',
'w^a-sil+xx=xx/A:xx+xx+xx/B:xx-xx_xx/C:xx_xx+xx/D:xx+xx_xx/E:5_5!0_xx-xx/F:xx_xx#xx_xx@xx_xx|xx_xx/G:xx_xx%xx_xx_xx/H:1_5/I:xx-xx@xx+xx&xx-xx|xx+xx/J:xx_xx/K:1+1-5']

Please check lab_format.pdf in HTS-demo_NIT-ATR503-M001.tar.bz2 for more details about full-context labels.

Grapheme-to-phoeneme (G2P)

In [1]: import pyopenjtalk

In [2]: pyopenjtalk.g2p("こんにちは")
Out[2]: 'k o N n i ch i w a'

In [3]: pyopenjtalk.g2p("こんにちは", kana=True)
Out[3]: 'コンニチワ'

Create/Apply user dictionary

1. Create a CSV file (e.g. user.csv) and write custom words like below:

ＧＮＵ,,,1,名詞,一般,*,*,*,*,ＧＮＵ,グヌー,グヌー,2/3,*

2. Call mecab_dict_index to compile the CSV file.

In [1]: import pyopenjtalk

In [2]: pyopenjtalk.mecab_dict_index("user.csv", "user.dic")
reading user.csv ... 1
emitting double-array: 100% |###########################################|

done!

3. Call update_global_jtalk_with_user_dict to apply the user dictionary.

In [3]: pyopenjtalk.g2p("GNU")
Out[3]: 'j i i e n u y u u'

In [4]: pyopenjtalk.update_global_jtalk_with_user_dict("user.dic")

In [5]: pyopenjtalk.g2p("GNU")
Out[5]: 'g u n u u'

About run_marine option

After v0.3.0, the run_marine option has been available for estimating the Japanese accent with the DNN-based method (see marine). If you want to use the feature, please install pyopenjtalk as below;

pip install pyopenjtalk[marine]

And then, you can use the option as the following examples;

In [1]: import pyopenjtalk

In [2]: x, sr = pyopenjtalk.tts("おめでとうございます", run_marine=True) # for TTS

In [3]: label = pyopenjtalk.extract_fullcontext("こんにちは", run_marine=True) # for text processing frontend only

LICENSE

• pyopenjtalk: MIT license (LICENSE.md)
• Open JTalk: Modified BSD license (COPYING)
• htsvoice in this repository: Please check pyopenjtalk/htsvoice/README.md.
• marine: Apache 2.0 license (LICENSE)

Acknowledgements

HTS Working Group for their dedicated efforts to develop and maintain Open JTalk.