kaljurand / inimesed

An Android app that lets you search your contacts by voice. Internet not required. Based on Pocketsphinx. Uses Estonian acoustic models.

Inimesed

Inimesed is an Android app that lets you search your contacts by voice. Since it uses the Android port of the Pocketsphinx speech recognizer, it does not require an internet connection. The current version of Inimesed uses Estonian acoustic models.

The FAQ tries to answer often occuring questions regarding porting Inimesed to other languages and use cases.

PocketSphinx on Android

In order to successfully build:

• PocketSphinx libraries must be installed
• swig must be installed
• Android NDK must be installed (tested with: android-ndk-r9)
• Android SDK must be installed
• change SPHINX_PATH in jni/Android.mk to match your configuration
  • pocketsphinx (PocketSphinx 0.7 / PocketSphinx v0.5.99)
  • sphinxbase (Sphinxbase-0.7)

Follow http://cmusphinx.sourceforge.net/2011/05/building-pocketsphinx-on-android/ to get everything installed.

Building for release

Run these commands in the app-directory. You need to have the files:

• local.properties
• speak.keystore
• assets/hmm/ should contain the acoustic models (see: https://github.com/alumae/et-pocketsphinx-tutorial/tree/master/models/hmm)

First time:

# Delete bin/ and gen/
ant clean

# This is needed only for the first time,
# and is run in jni-directory
cd jni/

# Create Java wrappers and pocketsphinx_wrap.c.
# They are already in the repository (generated using swig 2.0.4) so you do not
# necessarily need to run this.
# Note that swig 2.0.5+ breaks ndk_build. There is probably
# a simple solution to it but I haven't found it.
make swig_build

# Compile the c-files.
# First set an environment variable `ANDROID_NDK` to point to the Android NDK
make ndk_build

# Go back into the main directory
cd ..

# Compile and package everything for the release
# (asks for the release keys)
ant release

# Install
adb install -r bin/Inimesed-release.apk

Next time:

ant clean release
adb install -r bin/Inimesed-release.apk

Tags

Version tags are set by e.g.

git tag -a v0.8.18 -m 'version 0.8.18'

The last number should be even.

Lint

lint --html report.html .

inimesed's Issues

Update the Pocketsphinx engine

See: http://cmusphinx.sourceforge.net/wiki/tutorialandroid (code at https://github.com/cmusphinx/pocketsphinx-android-demo)

Things to try out (both supported by the latest Pocketsphinx):

• keyword search instead of grammar search
• hands-free mode, i.e. start listening automatically onResume + speech based ambiguity resolution

Use et-g2p for grapheme-to-phoneme conversion

• possibly with custom rules for foreign proper names
• measure the performance first

See https://github.com/alumae/et-g2p/

Add a boolean flag that determines if the SD card content (logs, grammars) is deleted onDestroy

Currently the logs and grammars are always deleted when the app is destroyed, which makes developing (e.g. discovering reasons for a crash) harder.

Running Demo

Hello I am trying to run the demo of ur project. it is crashing on mDecoder.startUtt();

I found the issue related to -jsgf config settings when I define the -lm file path it works fine. but -jsgf not working? I want to define my grammer. Can you tell me why defining -jsgf filepath not working??

Also it is giving error some:

08-12 17:33:27.887: I/dalvikvm(7804): Could not find method android.media.audiofx.AcousticEchoCanceler.create, referenced from method ee.ioc.phon.android.inimesed.SpeechAudioRecord.

These 3 lines making a problem in recording which causes crash

         c.setString("-hmm",  df.getHmm());
    c.setString("-dict", df.getDict());
    c.setString("-jsgf",   df.getJsgf());

When I replaced these with

        c.setString("-hmm","/sdcard/Android/data/edu.cmu.pocketsphinx/hmm/en_US/hub4wsj_sc_8k");
    c.setString("-dict","/sdcard/Android/data/edu.cmu.pocketsphinx/lm/en_US/hub4.5000.dic");
    c.setString("-lm","/sdcard/Android/data/edu.cmu.pocketsphinx/lm/en_US/hub4.5000.DMP");

it works fine than. Can you tell me what is the issue please??

Thanks

Add license