Singing Voice Synthesis System based on eSpeakNG

isengaara a3423be5a5 refactoring the dominant 6 سال پیش
.settings 507cb27df0 pull in the espeak activity sources from eyes-free 12 سال پیش
.tx e6c359dab9 Add a translation resource mapping for id => android/res/values-in (Android uses the old, deprecated code for Indonesia). 11 سال پیش
_layouts 0ad59a7f55 Support converting the markdown-based documentation to HTML. 9 سال پیش
android a3c1a00d45 android: fix downloading the android gradle plugin version 2.3.3. 6 سال پیش
dictsource 8d05fe2153 Fixes ssome foreign words 6 سال پیش
docs 8f30e3d607 Revert "en: use '@' instead of 'I' for the ROSES vowel in en and en-GB." 6 سال پیش
emscripten c98ed977f2 Adding IPA output to emscripten js library and demo 7 سال پیش
espeak-ng-data e20e767952 fr: add configuration files for MBROLA's Canadian French 1 and 2 voices 6 سال پیش
m4 dd3df1d0a8 Enable and fix -Wunused-function warnings. 7 سال پیش
phsource 19547ce75b ru: fix лёгких, мягкий, etc. 6 سال پیش
src a3423be5a5 refactoring the dominant 6 سال پیش
tests 9bf304b8ed tests: test mk consonant and vowel audio generation 6 سال پیش
tools 80ebc39d93 Update the emoji translations to CLDR 32. 7 سال پیش
vim 3057e8206d vim: support flag-only entries 7 سال پیش
.gitignore 1791e30e42 Add a fuzz target for SSML input 6 سال پیش
.travis.yml a72199f714 Run the tests as part of the Travis build. 7 سال پیش
CHANGELOG.md 35def0d760 Update the changelog. 6 سال پیش
COPYING c8526d17f8 autotools: add a COPYING file with the GPL v3 license to make autotools happy. 11 سال پیش
COPYING.APACHE eda3b394b6 Add the Apache 2.0 license (for the APK code in 'android'). 11 سال پیش
COPYING.BSD2 6514a475c7 Add information of the getopt.c BSD2 license to the README file. 8 سال پیش
COPYING.IEEE 547258e4bf Use an Open Source implementation of ieee80.c. 8 سال پیش
COPYING.UCD 4a5ec9a64f Include the Unicode license text in the main espeak-ng repository, as various places make use of Unicode data. 7 سال پیش
Makefile.am 137d56b272 refactoring part I 6 سال پیش
README.md 8022d12783 README: Wording changes based on suggestions by https://github.com/CarlosFromNewYork. 6 سال پیش
autogen.sh dffb80cb99 Added a CHANGELOG.md file. 8 سال پیش
configure.ac d7aee052c2 configure: Add a --disable-rpath option to prevent libtool hardcoding rpath variables. 6 سال پیش
espeak-ng.pc.in d5362b2608 build - Add pkg-config file 8 سال پیش
guix-env.scm 4eacd48ecf release new version 6 سال پیش
merge_github cc5ddbf857 will merge with github 6 سال پیش

README.md

eSpeak NG Text-to-Speech


The eSpeak NG (Next Generation) Text-to-Speech program is an open source speech synthesizer that supports 101 languages and accents, based on the eSpeak engine created by Jonathan Duddington. It supports spectral and Klatt formant synthesis, and the ability to use MBROLA voices.

See the CHANGELOG for a description of the changes in the various releases and with the eSpeak project.

The following platforms are supported:

Platform Minimum Version Status
Linux
BSD
Android 4.0
Windows Windows 8
Mac

Windows

Binaries

The Windows version of eSpeak NG 1.49.2 is available as:

You also need to install the Visual C++ Redistributable for Visual Studio 2015.

NOTE: SAPI 5 voices are not currently available in this release of eSpeak NG. There is an issue to track support for this feature.

Building

To build eSpeak NG on Windows, you will need:

  1. a copy of Visual Studio 2013 or later, such as the Community Edition;
  2. the Windows 8.1 SDK;
  3. the WiX installer plugin;
  4. the pcaudiolib project checked out to src (as src/pcaudiolib).

You can then open and build the src/windows/espeak-ng.sln solution in Visual Studio.

You can also use the Visual Studio tools to build espeak-ng with msbuild from the src/windows directory:

msbuild espeak-ng.sln

To build with a specific version of Visual Studio, you need to use:

msbuild /p:PlatformToolset=v120 espeak-ng.sln

replacing v120 with the appropriate value for the target Visual Studio version:

PlatformToolset Visual Studio
v120 2013
v140 2015
v141 (default) 2017

Linux, Mac, BSD

Dependencies

In order to build eSpeak NG, you need:

  1. a functional autotools system (make, autoconf, automake, libtool and pkg-config);
  2. a functional c compiler that supports C99 (e.g. gcc or clang).

Optionally, you need:

  1. the pcaudiolib development library to enable audio output;
  2. the sonic development library to enable sonic audio speed up support;
  3. the ronn man-page markdown processor to build the man pages.

To build the documentation, you need:

  1. the kramdown markdown processor.

On Debian-based systems such as Debian, Ubuntu and Mint, these dependencies can be installed using the following commands:

Dependency Install
autotools sudo apt-get install make autoconf automake libtool pkg-config
c99 compiler sudo apt-get install gcc
sonic sudo apt-get install libsonic-dev
ronn sudo apt-get install ruby-ronn
kramdown sudo apt-get install ruby-kramdown

Building

The first time you build eSpeak NG, or when you want to change how to build eSpeak NG, you need to run the following standard autotools commands:

./autogen.sh
./configure --prefix=/usr

NOTE: The --prefix option above will install the files to the /usr directory, instead of the default /usr/local location. You can use other standard configure options to control the output. For more information, you can run:

./configure --help

To use a different compiler, or compiler flags, you can specify these before the configure command. For example:

CC=clang-3.5 CFLAGS=-Wextra ./configure --prefix=/usr

The espeak-ng and speak-ng programs, along with the espeak-ng voices, can then be built with:

make

NOTE: Building the voice data does not work when using the -jN option. If you want to use that option, you can run:

make -j8 src/espeak-ng src/speak-ng
make

The documentation can be built by running:

make docs

Specific languages can be compiled by running:

make LANG

where LANG is the language code of the given language. More information can be found in the Adding or Improving a Language documentation.

Cross Compilation

Because the eSpeak NG build process uses the built program to compile the language and voice data, you need to build it locally first. Once you have built it locally you can perform the cross compilation using:

./configure --build=... --host=... --target=...
make -B src/espeak-ng src/speak-ng

Sanitizer Flag Configuration

It is possible to build eSpeak NG with the gcc or clang sanitizer by passing the appropriate CFLAGS and LDFLAGS options to configure. For example:

CFLAGS="-fsanitize=address,undefined -g" \
    LDFLAGS="-fsanitize=address,undefined" \
    CC=clang-6.0 ./configure
make
make check

NOTE: The -fsanitize=fuzzer option does not work when using the above configuration method. This is because clang will use the libFuzzer library which defines its own main and requires LLVMFuzzerTestOneInput to be defined. This breaks the autoconf check to see if the C compiler works.

LLVM Fuzzer Support

To enable libFuzzer support you need clang 6.0 or later. It is enabled with the following:

CC=clang-6.0 ./configure --with-libfuzzer=yes
make
make check

eSpeak NG Feature Configuration

The following configure options control which eSpeak NG features are enabled:

Option Description Default
--with-klatt Enable Klatt formant synthesis. yes
--with-mbrola Enable MBROLA voice support. yes
--with-sonic Use the sonic library to support higher WPM. yes
--with-async Enable asynchronous commands. yes

NOTE: The --with-sonic option requires that the sonic library and header is accessible on the system.

Extended Dictionary Configuration

The following configure options control which of the extended dictionary files to build:

Option Extended Dictionary Default
--with-extdict-ru Russian no
--with-extdict-zh Mandarin Chinese no
--with-extdict-zhy Cantonese no

The extended dictionaries are taken from http://espeak.sourceforge.net/data/ and provide better coverage for those languages, while increasing the resulting dictionary size.

Testing

Before installing, you can test the built espeak-ng using the following command from the top-level directory of this project:

ESPEAK_DATA_PATH=`pwd` LD_LIBRARY_PATH=src:${LD_LIBRARY_PATH} src/espeak-ng ...

The ESPEAK_DATA_PATH variable needs to be set to use the espeak-ng data from the source tree. Otherwise, espeak-ng will look in $(HOME) or /usr/share/espeak-ng-data.

The LD_LIBRARY_PATH is set as espeak uses the libespeak-ng.so shared library. This ensures that espeak uses the built shared library in the src directory and not the one on the system (which could be an older version).

Installing

You can install eSpeak NG by running the following command:

sudo make LIBDIR=/usr/lib/x86_64-linux-gnu install

NOTE: The LIBDIR path may be different to the one on your system (the above is for 64-bit Debian/Ubuntu releases that use the multi-arch package structure -- that is, Debian Wheezy or later).

You can find out where espeak-ng is installed to on your system if you already have an espeak-ng install by running:

find /usr/lib | grep libespeak-ng

Android

The espeak-ng sources contain the code for the Android™ port of the application. This is published as the eSpeak for Android program on the Google Play store. It is based on the eyes-free port of eSpeak to the Android platform. This code was originally maintained in a separate branch when the repository tracked eSpeak releases.

Dependencies

In order to build the Android APK file, you need:

  1. the Android SDK with API 21 support;
  2. the Android NDK;
  3. Gradle 2.1 or later.

In order to use Android Studio, you will also need:

  1. Android Studio.

Building with Gradle

  1. Set the location of the Android SDK:

    $ export ANDROID_HOME=<path-to-the-android-sdk>
    
  2. Build the project:

    $ ./autogen.sh
    $ ./configure --with-gradle=<path-to-gradle>
    $ make apk-release
    

This will create an android/build/outputs/apk/espeak-release-unsigned.apk file.

Signing the APK

In order to install the built APK you need to self-sign the package. You can do this by:

  1. Creating a certificate, if you do not already have one:

    $ keytool -genkey -keystore [YOUR_CERTIFICATE] -alias [ALIAS]
    
  2. Sign the package using your certificate:

    $ jarsigner -sigalg MD5withRSA -digestalg SHA1 \
      -keystore [YOUR_CERTIFICATE] \
      android/build/outputs/apk/espeak-release-unsigned.apk [ALIAS]
    
  3. Align the apk using the zipalign tool.

    $ zipalign 4 android/build/outputs/apk/espeak-release-unsigned.apk \
      android/build/outputs/apk/espeak-release-signed.apk
    

Installing

Now, you can install the APK using the adb tool:

$ adb install -r android/build/outputs/apk/espeak-release-signed.apk

After running, eSpeakActivity will extract the espeakdata.zip file into its own data directory to set up the available voices.

To enable eSpeak, you need to:

  1. go into the Android Text-to-Speech settings UI;
  2. enable eSpeak TTS in the Engines section;
  3. select eSpeak TTS as the default engine;
  4. use the Listen to an example option to check if everything is working.

Documentation

The main documentation for eSpeak NG provides more information on using and creating voices/languages for eSpeak NG.

The espeak-ng and speak-ng command-line documentation provide a reference of the different command-line options available to these commands with example usage.

eSpeak Compatibility

The espeak-ng binaries use the same command-line options as espeak, with several additions to provide new functionality from espeak-ng such as specifying the output audio device name to use. The build creates symlinks of espeak to espeak-ng, and speak to speak-ng.

The espeak speak_lib.h include file is located in espeak-ng/speak_lib.h with an optional symlink in espeak/speak_lib.h. This file contains the espeak 1.48.15 API, with a change to the ESPEAK_API macro to fix building on Windows and some minor changes to the documentation comments. This C API is API and ABI compatible with espeak.

The espeak-data data has been moved to espeak-ng-data to avoid conflicts with espeak. There have been various changes to the voice, dictionary and phoneme files that make them incompatible with espeak.

The espeak-ng project does not include the espeakedit program. It has moved the logic to build the dictionary, phoneme and intonation binary files into the libespeak-ng.so file that is accessible from the espeak-ng command line and C API.

Historical Versions

The historical branch contains the available older releases of the original eSpeak that are not contained in the subversion repository.

1.24.02 is the first version of eSpeak to appear in the subversion repository, but releases from 1.05 to 1.24 are available at http://sourceforge.net/projects/espeak/files/espeak/.

These early releases have been checked into the historical branch, with the 1.24.02 release as the last entry. This makes it possible to use the replace functionality of git to see the earlier history:

git replace 8d59235f 63c1c019

NOTE: The source releases contain the big_endian, espeak-edit, praat-mod, riskos, windows_dll and windows_sapi folders. These do not appear in the source repository until later releases, so have been excluded from the historical commits to align them better with the 1.24.02 source commit.

Feedback

Test latest development version at eSpeakNG online.

Report bugs to the espeak-ng issues page on GitHub.

Look at and subscribe to eSpeakNG mailing list to view and discuss other related topics.

License Information

eSpeak NG Text-to-Speech is released under the GPL version 3 or later license.

The ieee80.c implementation is taken directly from ToFromIEEE.c.txt which has been made available for use in Open Source applications per the license statement on http://www.realitypixels.com/turk/opensource/. The only modifications made to the code is to comment out the TEST_FP define to make it useable in the eSpeak NG library, and to fix compiler warnings.

The getopt.c compatibility implementation for getopt support on Windows is taken from the NetBSD getopt_long implementation, which is licensed under a 2-clause BSD license.

Android is a trademark of Google Inc.