IFCC-Feature-Extraction

Copyright Speech Information Processing Lab, IIT Hyderabad . All Rights Reserved.

If you use this code

please cite us

References

1. Karthika Vijayan, P R Reddy and K Sri Rama Murty, "Significance of analytic phase of speech signals in speaker verification," Speech Communication (Elsevier), Jun. 2016.

2. Shekhar Nayak, Saurabhchand Bhati and K Sri Rama Murty, "An Investigation into Instantaneous Frequency Estimation Methods for Improved Speech Recognition Features," in Proc. IEEE Global Conference on Signal and Information Processing (GlobalSIP), Nov. 14-16, 2017, Montreal, Canada.

3. Karthika Vijayan, Vinay Kumar and K. Sri Rama Murty, “Feature extraction from analytic phase of speech signals for speaker verification”, in INTERSPEECH, September 2014, Singapore, pp.1658-1662.

We are providing IFCC feature extraction in 2 languages

1.python

2.bash

Follow these instructions to extract IFCC features

Using Kaldi

Libraries to install

1.fftw3

2.libsnd

after installing the library files compile the code ( ./compile.sh)

for extracting IFCC features for .sph files

comput-ifcc.cpp

for extracting IFCC features for wav files

compute-ifcc_wav.scp

use make_ifcc.sh code to extract IFCC features

change the comput-ifcc.cpp file for changing the number of coefficients

Using Python

python IFCC_features_extract.py <data> <log> <output> <num_jobs>

example:

Where

data ----> data directory of wavfiles

make_ifcc ----> log directory for checking the progress of feature extraction

ifcc ----> output directory where the IFCC features will store

num_jobs ----> no of jobs

python IFCC_features_extract.py data/train make_ifcc ifcc 1

NOTE:-

if you want to change output format, you can change output file format in IFCC_features_run.py and number of jobs according to your system

wav.scp:

uttID channel wave_file_location