DeepSpeech JNI Library

This library acts as a layer between the DeepSpeech native library and the Java bindings. This library contains the definitions of the native functions declared as native in the DeepSpeech Java bindings.

Building the JNI library

The JNI library can be built for desktop devices and Android devices. If you want to build the JNI library for Android devices, check out the Readme file of the deepspeechandroid project.

There are two main configurations for the JNI library

• CPU

  • Available on desktop and Android

• Cuda

  • Only available on desktop devices

To build a given configuration, modify the value of the CUDA_CONFIGURATION variable of the CMakeLists.txt build file.

To build the cuda configuration go to the initial declaration of the variable in the build file and change it to either TRUE or FALSE depending on whether you want to build the CUDA or CPU configuration.

set(CUDA_CONFIGURATION FALSE) # CPU configuration

set(CUDA_CONFIGURATION TRUE) # CUDA configuration

If you are building the library for android, (Use Android Studio and also follow instructions of Readme of deepspeechandroid project) the CUDA_CONFIGURATION boolean is automatically forced off as the CUDA computing toolkit is not available for Android. (And of course Nvidia CUDA supporting GPUs for mobile devices duh)

Platform

Set the platform accordingly to your target build platform so that

$ENV{JAVA_HOME}/include/${PLATFORM}

points to the platform specific JNI headers.

Examples

set(PLATFORM linux)

set(PLATFORM win32)

set(PLATFORM darwin)

...

If your target build configuration is different to the configuration of the system, change $ENV{JAVA_HOME} to $SET_JAVA_HOME, a variable that needs to be set to the java home directory of a JDK for your target platform.

eg.

set(SET_JAVA_HOME "/usr/lib/jvm/java-8-oracle")

Placing DeepSpeech binaries

Make sure to place the libdeepspeech library for your target platform in ${CMAKE_SOURCE_DIR}/libs/${PLATFORM}.

When building for Desktop devices, the library must be named libdeepspeech_cpu or libdeepspeech_cuda depending on whether the binary is built to use the CPU or CUDA.

For android, place the ABI specific library inside ${CMAKE_SOURCE_DIR}/libs/android/${ANDROID_ABI}.

The libdeepspeech library is built for the following ABIs:

• arm64-v8a
• armeabi-v7a
• x86_64

The needed libdeepspeech library can be downloaded from the DeepSpeech releases page.

Running the build

To build the libdeepspeech-jni binary, you have to have CMake installed. Download CMake from here: CMake releases

After the project directory is setup correctly, the libdeepspeech-jni library can be build using the command.

cmake build .

Your binary will be placed in the CMake build directory. eg cmake-build-debug or cmake-build-release.