1

I wanted to try out PocketSphinxAndroidDemo. I thought I meticulously followed the tutorial on how to get it running here: https://sites.google.com/site/opiatefuchs/home/pocketsphinxandroiddemo

But I must be missing something. I get the following error when I do a run as as an android application from eclipse:

jni/pocketsphinx_wrap.c:760:26: fatal error: pocketsphinx.h: No such file or directory compilation terminated. make: * [obj/local/armeabi/objs-debug/pocketsphinx_jni/pocketsphinx_wrap.c] Error 1

I'm leaning towards maybe I did the ndk-build wrong - but it all seemed to work properly in the cygwin terminal.

I'm hoping someone can point me in the right direction. Thanks.

I've reviewed this previous post and it does not seem to contain an answer for me: http://sourceforge.net/p/cmusphinx/discussion/help/thread/8369b305/

Stepan1010
  • 3,136
  • 1
  • 16
  • 21
  • Reading more about voice recognition on mobile devices, I guess it is better to just stick with the platform's speech API (if they provide one) - e.g. on android http://android-developers.blogspot.com/2010/03/speech-input-api-for-android.html – Stepan1010 Aug 30 '13 at 16:55
  • you might have missed the finer details, such as using the tar.bz version instead of zip file, or using Swig 1.3 and not any other version. It has worked for me on Windows Xp SP 3 and the speech detection is fairly accurate. Where did you get the `pocketsphinxandroiddemo` android project? I could not download it from the link provided in that tutorial and so I had to download it from another source, but then I had to make lots of changes in the tabs mentioned in step 9 of that tutorial. I haven't seen your particular error message, but perhaps it's a similar issue? – user13267 Oct 31 '13 at 09:50
  • The platform's speech API is very accurate, but it is available offline only in the latest android version and only supported in a few devices. For older phones and Android versions, it requires network connection. See my post here: http://stackoverflow.com/questions/19564561/does-the-google-speech-recognizer-on-android-need-internet – user13267 Oct 31 '13 at 09:51
  • @user13267 I actually did get it working and it was due to a missed step in the tutorial - someone should review that tutorial and rewrite it - some things have changed since it was written it appears. It was not very accurate on my phone (galaxy s3) - at least not compared to the Android speech API. I did not do any fine tuning so it might have had trouble with the microphone on the galaxy s3 - it does not have a great microphone. – Stepan1010 Oct 31 '13 at 13:54
  • I think the accuracy of pocketsphinx is not only related to the quality of the microphone but also to the tuning of the language model and dictionary used. They are included in the files that we need to copy in to the folder structure of the android phone. The steps to do this arr given in the CMU Sphinx wiki site. I don't remember exactly where, but the link to the wiki is somewhere in this post: http://stackoverflow.com/questions/19313316/why-does-pocketsphinx-android-demo-require-cygwin . But training the language model and dictionary will probably take a significant amount of time. – user13267 Nov 01 '13 at 01:12

0 Answers0