This package is deprecated and is no more maintained, please use shennong instead: https://github.com/bootphon/shennong

Installation:

In short if all dependences are installed on the system you can install feature_extraction within the cli:

python setup.py build && python setup.py install

If you are installing from source using conda and dependences are not installed you can do:

Gget feature_extraction from github:

>> git clone https://github.com/bootphon/features_extraction

Create your environment, for example if using conda (using the python 2.7 64-BIT from https://www.continuum.io/downloads)

>> cd features_extraction
>> conda create --name feat --file requirements.txt
>> source activate feat

Install other dependences that are not available with conda

(feat) >> pip install oct2py   
(feat) >> pip install git+http://github.com/bootphon/spectral
(feat) >> pip install git+https://github.com/bootphon/h5features    

If you are working from the source code, you will need to include features_extraction in your path, in bash/linux:

>> export PATH=$PATH:[DIR_FEATURES_EXTRACTION]
>> export PYTHONPATH=$PYTHONPATH:[DIR_FEATURES_EXTRACTION]
>> chmod 755 features.py

Where [DIR_FEATURES_EXTRACTION] is the directory where feature_extraction source code is keep.

Dependecies

python dependences:

• spectral
• h5features
• octave

octave dependences:

• ltfat
• amtoolbox
• rastamat
• signal (octave package), require control and general
• AuditoryToolbox

To run rastama the previous packages you will need a working octave's distribution, these packages depends of the octaves packages: general, control and signal. To install the packages, from octave's console line you'll need to run:

>> pkg install -forge general
>> pkg install -forge control
>> pkg install -forge signal

AuditoryToolbox depends on compiled octave/matlab mex files, to create these file you will need to compile the files within the AuditoryToolbox directory, in linux command line on AuditoryToolbox/src directory do:

make

Notes

for now, octave features dont have many options since they are hard coded. .mat output for python features not implemented yet center_times in some octave features are wrong...

Example:

python features.py test/wavs/*.wav -h5 my_mfccs.h5f -c mfcc_config.json

see config options at config_doc.rst