adore-hrzz/nao-sound-classification

Sound classification module, distinguishing between articulated speech and other sounds.

nao-sound-classification's Issues

Add yaml file parsing in C++ code

In order to avoid having to generate two separate listings of the labeled dataset.

Expand the dataset to make it more balanced

Use data from the homebank project.

Enable classification on prerecorded (.wav) files

Modify SCModule + autoClassify to work on prerecorded files.

Simplify and document learning procedure

Reduce the number of pre-processing and post-processing steps. Implement script options using the Python argparse library.

Classifier crash in noise filter

The module seems to be crashing on the robot. Also, apparently the module leaks memory.
@robotiks , @fpetric and @stribor14 please provide relevant details.

Examine failed learning examples

Examine the misclassified data in order to gain insights into possible causes of failure.

Clean up CPP code

Remove all functionality that can be better implmented in Python. Re-implement in Python.

Add validation scripts

Add Python scripts used for dataset validation

Check corelation between MFCC and BFCC

How strongly are these features correlated?

Update docs for classification

Currently, only the docs for learning and batch classification have been updated.

Docs for getting the classifier running, both as a local and as a remote module must be updated too!

Consider more advanced features from the Oller et al. 2010 paper

Consider extracting more advanced features, as described in the Oller et al. 2010.

Re-evaluate the loudness feature

Examine this feature on several samples. Consider how the feature is affected by de/amplifying the signal.

Upgrade to naoqi 2.1

Requires updates to the build system and maybe some minor changes to code.

Evaluate openSMILE for feature extraction

There is a know bug with LibXtract when extracting some features (@stribor14 which features exactly?) Try using openSMILE instead.

Crash in Learner

When running learner with the attached configuration files: SC.txt, test_data.txt, train_data.txt

learner crashes (after processing an number of files), with the following error message:

4month_18weeks_old_baby_cooing_and_babbling__nonarticulated_006.wav: [||||||||||] DONE
sto_veliki_misle_mali_govore_sta_je_kisobran__speech_031.wav: [||||||||||] DONE
OpenCV Error: Internal error (data is empty) in CvMLData::set_response_idx, file /home/damjan/devel/opencv/opencv-2.4.9/modules/ml/src/data.cpp, line 347
terminate called after throwing an instance of 'cv::Exception'
  what():  /home/damjan/devel/opencv/opencv-2.4.9/modules/ml/src/data.cpp:347: error: (-3) data is empty in function CvMLData::set_response_idx

Aborted (core dumped)

This is the part of the code where things fail (as reported by gdb):

    CvMLData modelData;
    modelData.set_miss_ch('@');
    modelData.read_csv((par.featData).c_str());
    
    modelData.set_response_idx(0);

@stribor14 any ideas?

Clean up folder layout

Move source files to src folder and headers to include folder. Requires modifications to CMakeLists.txt