Current tutorial describes a most recent verison pocketsphinx-android-demo-5prealpha which is available in downloads
You should have ADT plugin installed. Also will need Android SDK installed either manually or with ADT plugin.
ant debug install
It will build and install application signed with a debug key. To build and install a release version run
ant release install
Put the obtained jar (the one without ”-nolib” suffix) into “libs” directory of your Android project. It should be automatically referenced either by a building script or by IDE - depending on what you are using. Currently the only officially support IDE-s are Android Studio and Eclipse (see below).
Due to a nasty bug in ADT plugin it is currently impossible to include pocketsphinx-android as a single jar. You will have to use ”-nolib” version of the library and add shared libraries manually. For example, if you are building application for ARM architecture, you will have “libs/pocketsphinx-android-0.8-nolib.jar” and “libs/armeabi/libpocketsphinx_jni.so”.
The standard way to ship resource files with your application in Android is to put them in
assets/ directory of your project. But in order to make them available for pocketsphinx files should have physical path, as long as they are within .apk they don't have one. Assets class from pocketsphinx-android provides a method to automatically copy asset files to external storage of the target device.
edu.cmu.pocketsphinx.Assets#syncAssets synchronizes resources reading items from
assets.lst file located on the top
assets/. Before copying it matches MD5 checksums of an asset and a file on external storage with the same name if such exists. It only does actualy copying if there is incomplete information (no file on external storage, no any of two .md5 files) or there is hash mismatch. PocketSphinxAndroidDemo contains
ant script that generates
assets.lst as well as
.md5 files, look for
Please note that if ant build script doesn't run properly in your build process, assets might be out of sync. Make sure that script runs or create md5 files and assets.lst yourself.
The classes and methods of pocketsphinx-android were designed to resemble the same workflow used in pocketsphinx, except that basic data structures organized into classes and functions working with them are turned into methods of the corresponding classes. So if you are familiar with pocketsphinx you should feel comfortable with pocketsphinx-android too.
SpeechRecognizer is the main class to access decoder functionality. It is created with the help of
SpeechRecognizerBuilder allows to configure main properties as well as other parameters of teh decoder. The parameters keys and values are the same as those are passed in command-line to pocketsphinx binaries. Read more about tweaking pocketsphinx performance.
File modelsDir = new File(assetsDir, "models"); recognizer = defaultSetup() .setAcousticModel(new File(modelsDir, "hmm/en-us-semi")) .setDictionary(new File(modelsDir, "dict/cmu07a.dic")) .setRawLogDir(assetsDir).setKeywordThreshold(1e-20f) .getRecognizer(); recognizer.addListener(this);
Decoder configuration is lengthy process that contains IO operation, so it's recommended to run in inside async task.
Decoder supports multiple named searches which you can switch in runtime
// Create keyword-activation search. recognizer.addKeyphraseSearch(KWS_SEARCH, KEYPHRASE); // Create grammar-based searches. File menuGrammar = new File(modelsDir, "grammar/menu.gram"); recognizer.addGrammarSearch(MENU_SEARCH, menuGrammar); File digitsGrammar = new File(modelsDir, "grammar/digits.gram"); recognizer.addGrammarSearch(DIGITS_SEARCH, digitsGrammar); // Create language model search. File languageModel = new File(modelsDir, "lm/weather.dmp"); recognizer.addNgramSearch(FORECAST_SEARCH, languageModel);
Once you setup the decoder and add all the searches you can start recognition with
You will get notified on speech end event in
onEndOfSpeech callback of the recognizer listener. Then you could call
recognizer.cancel(). Latter will cancel the recognition, former will cause the final result
be passed you in
During the recognition you will get partial results in
You can also access other Pocketsphinx method wrapped with Java classes in swig, check for details Decoder, Hypothesis, Segment and NBest classes.
Pocketsphinx is provided with prebuilt binaries and it's not easy to compile it on various platforms. You shouldn't build it unless you understand what you are doing. Use prebuilt binaries instead.
Make sure you have “pocketsphinx-android” in the same parent directory as “pocketsphinx” and “sphinxbase” are as the building script refers them using exactly these names. If done right you will find multiple shared libraries in “lib” - each in the directory with the target platform name, and two jars in “build”.