This progress report summarizes the work done so far on developing a dictionary learning-based spoken language identification system. It discusses the objectives of designing efficient dictionary learning algorithms and extracting suitable parameters. It reviews literature on different dictionary learning algorithms like K-SVD, LC-KSVD, D-KSVD, and Discriminative Bayesian dictionary learning. The report explains the working of these algorithms and highlights progress made in implementing MATLAB code for an existing algorithm and extracting/analyzing parameters to compare algorithm outputs and identify the most efficient one. Future work involves completing the MATLAB code and experiments.
Mini project progress on spoken language ID system
1. MINI PROJECT PROGRESS REPORT ON
“DEVELOPMENT OF LEARNED DICTIONARY BASED
SPOKEN LANGUAGE IDENTIFICATION SYSTEM”
Under the guidance of
Mr. Om Prakash Singh
Associate Professor
ECE Dept., SMIT, Majitar.
Presented By:
PALLAVI BHARTI
(20130549) 7thsem.
RICHA BHARTI
(20130707) 7thsem.
2. Objective:
Design and Analysis of Efficient Algorithms for Dictionary learning.
Extraction of suitable parameters for efficient calculation.
Mathematical modelling of an efficient algorithm for dictionary
learning.
3. Planning:
Perform statistical analysis of pre-existing algorithm.
Develop MATLAB code for algorithm design application.
Design of algorithm for language identification.
4. Progress:
Literature Survey about:
1. What is dictionary learning?
2. Different types of Algorithms for dictionary learning.
K-SVD.
LC-KSVD.
D-KSVD.
Discriminative Bayesian dictionary learning.
5. What is Dictionary learning?
Dictionary learning is compact representation of training data or test
data .it is the best way to represent a signal.
Dictionary learning is a branch of signal processing that aims at finding
a frame (called dictionary) in which some training data admits sparse
representation.
6. K-SVD Algorithm:
1. K-means clustering process for adapting dictionaries in order to achieve
sparse signal representation.
2. Application that can benefit from the sparsity and over completeness
concepts include compression and feature extraction.
3. In order to over complete and sparse representation the simple
technique is Orthogonal matching pursuit(OMP).
7. D-KSVD Algorithm:
1. Discriminative K-SVD (D-KSVD), a newly proposed dictionary learning
method, has better discrimination ability since it incorporates the
classification error into its object function and learns a
discriminative dictionary and a linear classifier .
2. D-KSVD is a two-step iterative method that is intialiation and
classification, and its convergence speed is heavily influenced by the
initialization values, initialization method is proposed for the D-KSVD
dictionary learning algorithm.
3 Naive Bayesian classifier is used to initialize the linear classifier in D-
KSVD.
8. LC-KSVD Algorithm:
1. A label consistent K-SVD (LC-KSVD) algorithm to learn a
discriminative dictionary for sparse coding is presented.
2. In addition to using class labels of training data, we also associate
label information with each dictionary item (columns of the
dictionary matrix) to enforce discriminability in sparse codes during
the dictionary learning process.
3. Algorithm learns a single over-complete dictionary and an optimal
linear classifier jointly. It yields dictionaries so that feature points
with the same class labels have similar sparse codes.
9. Discriminative Bayesian dictionary
learning:
1. The proposed approach infers the probablity distribuition over the
atoms of a discriminative dictionary using a finite approximation of
beta process.
2. It also computes sets of bernoulli distribuition that associate a class
labels to the learned dictionary atoms.
3. Test instance is first sparsely encoded over the learned dictionary
and the codes are fed to the classifier
11. Jobs to be done:
Implementation of MATLAB code for pre-existing algorithm.
Study of the corresponding pre-existing algorithms.
Extraction of the parameters and then observation and analysis for
output.
The result will be compared and plotted in the code set prepared in
MATLAB 2011.
Checking and comparison of the output for most efficient algorithm
design.
12. References:
[1] K. Engan, S. Aase, and J. Husφy. Frame based signal compression using method
of optimal directions (mod), 1999. IEEE Intern. Symp. Circ. Syst., 1999.
[2] D. Pham and S. Venkatesh. Joint learning and dictionary construction for
pattern recognition, 2008. CVPR.
[3] Q. Zhang and B. Li. Discriminative k-svd for dictionary learning in face
recognition, 2010. CVPR.
[4] "Discriminative Bayesian Dictionary Learning for Classification", IEEE
Transactions on Pattern Analysis & Machine Intelligence, , no. 1, pp. 1, PrePrints
PrePrints, doi:10.1109/TPAMI.2016.2527652