OpenJTalk的Python包装器

导航

已验证详情

这些详情已由PyPI验证
维护者
Avatar for syoyo from gravatar.com syoyo

未验证详情

这些详情尚未由PyPI验证
项目链接
元数据
  • 许可证: MIT许可证 (MIT)
  • 作者: Ryuichi Yamamoto
  • 标签 OpenJTalk, 研究
分类

项目描述

pyopenjtalk

PyPI Python package Build Status License DOI

OpenJTalk的Python包装器。

该包包含两个核心组件

  • 基于OpenJTalk的文本处理前端
  • 使用HTSEngine的语音合成后端

注意

  • 该包使用OpenJTalk的修改版本构建。修改版本提供了相同的功能和一些改进（例如，cmake支持），但与HTS工作组的不同。
  • 该包还使用hts_engine_API的修改版本。同样适用。

在使用pyopenjtalk包之前，请查看两个软件的LICENSE。

构建要求

该Python包依赖于cython来为open_jtalk和hts_engine_API创建Python绑定。您必须需要以下工具来构建和安装pyopenjtalk

  • C/C++编译器（用于构建C/C++扩展）
  • cmake
  • cython

支持的平台

  • Linux
  • Mac OSX
  • Windows (MSVC)（见此PR）

安装

pip install pyopenjtalk

开发

要本地构建包，请确保克隆open_jtalk和hts_engine_API。

git submodule update --recursive --init

然后运行

pip install -e .

快速演示

请在此处查看笔记本版本（nbviewer）。

文本到语音（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))

仅运行文本处理前端

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']

请检查 lab_format.pdf 文件，它位于 HTS-demo_NIT-ATR503-M001.tar.bz2 中，获取关于全上下文标签的更多详细信息。

字符到音素 (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'

关于 run_marine 选项

自 v0.3.0 版本起，run_marine 选项已可用于使用基于 DNN 的方法估计日语口音（参见 marine）。如果您想使用此功能，请按照以下步骤安装 pyopenjtalk：

pip install pyopenjtalk[marine]

然后，您可以使用以下示例使用该选项；

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

许可证

致谢

HTS 工作组对开发和维护 Open JTalk 的专注努力表示敬意。

由以下支持