OpenJTalk on Windows10 (Speak Japanese with Python from environment construction)

Introduction

Windows10 it was to try the OpenJTalk in, but the necessary environment to be for anyone because it was trimmed from scratch Make a note. As the title says, I will explain from environment construction to how to speak Japanese with Python.

environment

• windows 10
• Visual Studio 2019 Community
• OpenJTalk 1.09
• Python 3.7

Build environment preparation

On Windows, you need to build it yourself to run OpenJTalk, so prepare the environment.

Installing Visual Studio 2019 Community

This is not the main subject, so I will save it.

Build settings

Start the command prompt and execute the following two commands.

cd C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Auxiliary\Build

vcvarsall.bat x64

nmake settings

If you try to execute the nmake command at the command prompt and you get an error that the command cannot be executed (an error that the command is not recognized), you need to put the path of nmake.exe in the environment variable. (I had to pass it) I was able to take the following pass. C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.24.28314\bin\Hostx64\x64

Install OpenJTalk

From here, I referred to the site of here. Thank you.

Obtaining source code OpenJTalk, hts_engineAPI

First, get OpenJTalk 1.09. Select from here, but it will be downloaded immediately at the link below.

http://downloads.sourceforge.net/open-jtalk/open_jtalk-1.09.tar.gz

Select the speech synthesis library hts_engineAPI required for compilation from here, but like OpenJTalk, it will be downloaded immediately with the following link.

http://downloads.sourceforge.net/hts-engine/hts_engine_API-1.10.tar.gz

Extract the downloaded ʻopen_jtalk-1.09.tar.gz Set it toc: \ temp \ open_jtalk-1.09. Also, if the hts_engineAPI is expanded to c: \ temp \ open_jtalk-1.09 \ hts_engine_API-1.10`, problems will be reduced when compiling.

Compile with nmake

Now that you can use nmake at the command prompt, compile it.

Compiling hts_engine_API-1.10.

First, start from hts_engine_API-1.10, so execute the following three commands. When completed, a file will be created in c: \ hts_engine_API.

cd c:\temp\open_jtalk-1.09\hts_engine_API-1.10
nmake /f Makefile.mak
nmake /f Makefile.mak install

Compile open_jtalk-1.09

Next, compile open_jtalk-1.09. Execute the following three commands. When completed, a file will be created in c: \ open_jtalk.

cd c:\temp\open_jtalk-1.09
nmake /f Makefile.mak
nmake /f Makefile.mak install

A garbled character string is displayed near the end, but it seems that you do not have to worry about it for the time being. Click here for details (https://qiita.com/kkoba84/items/b828229c374a249965a9)

This will generate c: \ open_jtalk \ bin \ open_jtalk.exe.

Get Voice data

From MMDAgent https://sourceforge.net/projects/mmdagent/files/MMDAgent_Example/MMDAgent_Example-1.6/ Download MMDAgent_Example-1.6.zip from, and copy Voice / mei / *. Htvoice to c: \ open_jtalk \ bin . It is the data of four emotions of a girl named mei.

This completes the preparation.

Open JTalk execution test

Assuming that a file with Japanese input is created as input.txt in c: \ open_jtalk \ bin ， Execute the following command.

cd c:\open\jtalk\bin
open_jtalk -m mei_normal.htsvoice -x ../dic -ow output.wav input.txt

If output.wav is created in the same directory, it is successful. Try double-clicking it to see if it plays. At this point, it may not work in Japanese due to the influence of encoding. Actually I didn't work.

Run OpenJTalk from Python

Use the winsound module to run OpenJTalk from Python on Windows. Also, since the dictionary generation in this compilation on Windows was shift-jis, it is necessary to convert the encoding when passing from the internal encoding of python to stdin.

jtalk.py

#coding: utf-8
import subprocess
from datetime import datetime
import winsound

def jtalk(t):
    # depend on your install folder
    OPENJTALK_BINPATH = 'c:/open_jtalk/bin'
    OPENJTALK_DICPATH = 'c:/open_jtalk/dic'
    # VOICEPATH -> can change mei voice
    OPENJTALK_VOICEPATH = 'c:/open_jtalk/bin/mei_normal.htsvoice'
    open_jtalk=[OPENJTALK_BINPATH + '/open_jtalk.exe']
    mech=['-x',OPENJTALK_DICPATH]
    htsvoice=['-m',OPENJTALK_VOICEPATH]
    speed=['-r','1.0']
    outwav=['-ow','open_jtalk.wav']
    cmd=open_jtalk+mech+htsvoice+speed+outwav
    c = subprocess.Popen(cmd,stdin=subprocess.PIPE)

    # convert text encoding from utf-8 to shitf-jis
    c.stdin.write(t.encode('shift-jis'))
    c.stdin.close()
    c.wait()

    # play wav audio file with winsound module
    winsound.PlaySound('open_jtalk.wav', winsound.SND_FILENAME)


def say_datetime():
    # get datetime and call jtalk
    d = datetime.now()
    text = '%s month%s day,%s time%s minutes%s seconds' % (d.month, d.day, d.hour, d.minute, d.second)
    jtalk(text)

if __name__ == '__main__':
    say_datetime()

Place it in any directory and start it with Python jtalk.py. Of course, the startup method is free. It is successful if the computer speaks the date and time.

Use as a Python module

Call the above file (jtalk in jtalk.py) from the same directory (also free here). Also, since the argument is the character you want to speak, specify it. ʻU''` also specifies the encoding.

test.py

import jtalk
jtalk.jtalk(u'That's it')

that's all. It's been a long time, but thank you!

The site that I used as a reference

https://qiita.com/Gaccho/items/f748f59784ff68d7a474 https://qiita.com/kkoba84/items/b828229c374a249965a9