I have a working model built with HTK for online handwriting recognition. Now I want to build another recognition system for Android. After some research I found out that PocketSphinx has a version running on Android OS. Now, what I think of is to convert my HTK model to Sphinx ( or to rebuild it in Sphinx) and then use it with PocketSphinx-on-Android.
First, is there any point that I miss in my plan?
Second, will the converted system (or rebuild with Sphinx) be directly useable in PocketSphinx? In other words, will there be any further conversion/adaptation for a Sphinx model to PocketSphinx?
Thanks in advance
Now, what I think of is to convert my HTK model to Sphinx ( or to rebuild it in Sphinx) and then use it with PocketSphinx-on-Android.
Pocketsphinx has more strict format of HMM which is more speech-oriented. For example, all HMMs must have equal number of states. So it might not be easy to convert. It is easier to retrain.
Second, will the converted system (or rebuild with Sphinx) be directly useable in PocketSphinx? In other words, will there be any further conversion/adaptation for a Sphinx model to PocketSphinx?
Model conversion is not required.
However, there is an issue that pocketsphinx on android allows only voice input, it has no means to input custom features. So you will have to implement that part yourself.
Related
Has anyone ever programmed using CNTK for reading hand-filled documents? I tried OCRs and they dont do handwriting recognition at all (next to nothing). Thinking of using CNTK for the same. I searched and found that not many have tried such a thing. Any advice on libraries or any pointers anyone?
Here a basic OCR example using CNTK:
https://github.com/Microsoft/CNTK/blob/master/Tutorials/CNTK_103B_MNIST_FeedForwardNetwork.ipynb
However, in order to use the model in a real application you will need a way to segment the handwritting.
Is it possible to adapt an acoustic model using "sphinx4" only? I have already checked the website but the commands mentioned are for pocketShphinx "Adapting the default acoustic model".
I have also checked some other solutions but all those used pocketsphinx.
It is not possible, you need sphinxtrain and pocketsphinx for adaptation.
While trying to use OpenCV for face detection on Windows, I need to pull in almost all the libraries (2d, 3d, ml, gui etc.). Otherwise my program wouldn't run. I am not really sure why I need any GUI for something as algorithmic as object detection. What is the minimal set of libraries required and is there a special way to build OpenCV such that there aren't that many dependencies?
You need opencv_core to get base objects like cv::Mat, opencv_imgproc to use thresholds, histograms and other image pre-processing, and opencv_highgui for reading, writing and displaying images, and using video streams from cameras and video files. That's all I can tell you without knowing how to run openCV on Windows, and not knowing which version of openCV You are using. As far as I know there is no way of building only some parts of openCV.
Generally from my experience You only need to add libraries associated to headers which You are using. So, if you have problems with tracking them try to avoid using #include "opencv2/opencv.hpp" and try a bit harder way of #include "opencv2/core/core.hpp" etc.
Yes, you can build OpenCV without certain library features. OpenCV uses CMake, which requires a little learning if you don't know it already, but you can uncheck OpenCV components you don't need in the CMake build configuration.
You can get away without using highgui in your app if you can read images with some other library (but not sure if you can build OpenCV without it).
Also - you will need to #include "opencv2/objdetect/objdetect.hpp" for support of Haar cascade classifiers (as of OpenCV 2.3.1).
I'm trying to find some free XNA friendly FBX Models to use in my test application. I found some 3DS models and converted them to FBX and tried using them, but I just see some 'lines' on the screen .. I know I may have to scale/translate them, but I tried doing this and nothing worked ... Some models won't show anything at all, not even the line artifacts ..
Any ideas where I can get fairly simply FBX models from which can be used in XNA 3.1 ?
Also, there are some models I saw which contain mutiple FBX files .. I dont want these as they require more complexity .. Right now I'm just looking for models contained within one FBX file ..
http://turbosquid.com is a great place to get some free models, but the paid ones are always better.
http://www.mixamo.com/ also has some free character models. No animations though, unless you pay. However, if you're looking for animations, Carnegie mellon university has a whole slew of motion-capture files for free. You'd have to re-rig the character yourself using 3ds max or whatever 3D program you like to use, so that they use "biped" instead of bones, that's the only way to use BVH.
Motion capture BVH files can be found here: https://sites.google.com/a/cgspeed.com/cgspeed/motion-capture/cmu-bvh-conversion
I need to program a util for a 3d model. What i need to do is pass the 3d model to the util and have it extract multiple information. Things like poly count, size/scale (if applicable) and anything else i can grab. Then i need to take a screen shot of the model 8 times (45deg from 0 to 360). Maybe 16. Is there an app that i can use to extract data from the model and is there another app i can use to create the image? if so i can write a php script to do this all for me :) if not i'll post more question about specifics.
My OS is debian etch
the models will be any format. I may make it 3ds only or limit it to what the app(s) support. Or i may have another app to convert one format to the other and extract the data from that format.
I haven't tried this, but you could try 3ds2pov to convert your 3DS files into POV-Ray format, and then render with POV-Ray.
The 3ds2pov program is quite old, so I don't know how compatible it is with current 3DS files. The archive comes with source code so ought to build without too much difficulty on Debian.
POV-Ray itself runs easily on Linux.
nb: other 3DS to POV converters may exist. This one just happens to be the first one returned by Googling for "3ds to pov".
Your requirements are very loose and vague but most 3d engines should have the tools you need to create your custom utility. I would start by looking at Blender or the viewer tools from OGRE, OpensceneGraph and similar tools.
You can try this: Linux based A3dsViewer -> does 3ds to pov conversion.