Hamzeh Ghasemzadeh

Dr. Hamzeh Ghasemzadeh

Michigan State University
orcid

Research Focus & Metrics

Academic Career Length

14Years (2013-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
11 March 2024 in Journal of Speech, Language, and Hearing Research
Verified
Journal Article
Closed access
HGHamzeh Ghasemzadeh
Hamzeh Ghasemzadeh
RHRobert E. Hillman
Robert E. Hillman
DMDaryush Mehta
Daryush Mehta
Purpose: Many studies using machine learning (ML) in speech, language, and hearing sciences rely upon cross-validations with single data splitting. This study's first purpose is to provide quantitative evidence that would incentivize researchers to instead use the more robust data splitting method of nested k -fold cross-validation. The second purpose is to present methods and MATLAB code to perform power analysis for ML-based analysis during the design of a study. Method: First, the significant impact of different cross-validations on ML outcomes was demonstrated using real-world clinical data. Then, Monte Carlo simulations were used to quantify the interactions among the employed cross-validation method, the discriminative power of features, the dimensionality of the feature space, the dimensionality of the model, and the sample size. Four different cross-validation methods (single holdout, 10-fold, train–validation–test, and nested 10-fold) were compared based on the statistical power and confidence of the resulting ML models. Distributions of the null and alternative hypotheses were used to determine the minimum required sample size for obtaining a statistically significant outcome (5% significance) with 80% power. Statistical confidence of the model was defined as the probability of correct features being selected for inclusion in the final model. Results: ML models generated based on the single holdout method had very low statistical power and confidence, leading to overestimation of classification accuracy. Conversely, the nested 10-fold cross-validation method resulted in the highest statistical confidence and power while also providing an unbiased estimate of accuracy. The required sample size using the single holdout method could be 50% higher than what would be needed if nested k -fold cross-validation were used. Statistical confidence in the model based on nested k -fold cross-validation was as much as four times higher than the confidence obtained with the single holdout–based model. A computational model, MATLAB code, and lookup tables are provided to assist researchers with estimating the minimum sample size needed during study design. Conclusion: The adoption of nested k -fold cross-validation is critical for unbiased and robust ML studies in the speech, language, and hearing sciences. Supplemental Material: https://doi.org/10.23641/asha.25237045

1 April 2016 in Digital Signal Processing
Verified
Journal Article
Closed access
HGHamzeh Ghasemzadeh
Hamzeh Ghasemzadeh
MKMehdi Tajik Khass
Mehdi Tajik Khass
MAMeisam Khalil Arjmandi
Meisam Khalil Arjmandi

1 September 2015 in Biomedical Signal Processing and Control
Verified
Journal Article
Closed access
HGHamzeh Ghasemzadeh
Hamzeh Ghasemzadeh
MKMehdi Tajik Khass
Mehdi Tajik Khass
MAMeisam Khalil Arjmandi
Meisam Khalil Arjmandi
MPMohammad Pooyan
Mohammad Pooyan

1 August 2018 in IET Signal Processing
Verified
Journal Article
Closed access
HGHamzeh Ghasemzadeh
Hamzeh Ghasemzadeh
MKMohammad H. Kayvanrad
Mohammad H. Kayvanrad
Recently, merging signal processing techniques with information security services has drawn a lot of attention. Steganography and steganalysis are among those trends. Like their counterparts in cryptology, steganography and steganalysis are in a constant battle. Steganography methods try to hide the presence of covert messages in innocuous‐looking data, whereas steganalysis methods try to break steganography algorithms and reveal the existence of hidden messages. The streaming nature of audio signals, their popularity, and their wide spread usage make them very good candidates for steganography. This has led to a very rich literature on both steganography and steganalysis of audio signals. This study intends to conduct a comprehensive review of audio steganalysis methods aggregated over near 15 years. To that end, both compressed and con‐compressed methods are reviewed, and then their important details are presented in different tables. Furthermore, some of the most recent audio steganalysis methods (both non‐compressed and compressed ones) are implemented and comparative analyses on their performances are conducted. Finally, the study provides some possible directions for future research studies on audio steganalysis.

1 October 2017 in IET Signal Processing
Verified
Journal Article
Open access
PDF file
HGHamzeh Ghasemzadeh
Hamzeh Ghasemzadeh
MAMeisam Khalil Arjmandi
Meisam Khalil Arjmandi
Calibration and higher‐order statistics are standard components of image steganalysis. However, these techniques have not yet found adequate attention in audio steganalysis. Specifically, most of current studies are either non‐calibrated or only based on noise removal. The goal of this study is to fill these gaps and to show that calibrated features based on re‐embedding technique improve performance of audio steganalysis. Furthermore, the authors show that least significant bit is the most sensitive bit plane to data hiding algorithms, and therefore it can be employed as a universal embedding method. The proposed features also benefit from an efficient model which is tailored to the needs for audio steganalysis and represent the maximum deviation from human auditory system. Performance of the proposed method is evaluated on a wide range of data hiding algorithms in both targeted and universal paradigms. The results show the effectiveness of the proposed method in detecting the finest traces of data hiding algorithms in very low embedding rates. The system detects Steghide at capacity of 0.06 bit per symbol with sensitivity of 98.6% (music) and 78.5% (speech). These figures are, respectively, 7.1% and 27.5% higher than the state‐of‐the‐art results based on R‐Mel‐frequency cepstral coefficient features.

1 November 2020 in Journal of Voice
Verified
Journal Article
Closed access
HGHamzeh Ghasemzadeh
Hamzeh Ghasemzadeh
DDDimitar D. Deliyski
Dimitar D. Deliyski
DFDavid S. Ford
David S. Ford
JKJames B. Kobler
James B. Kobler
RHRobert E. Hillman
Robert E. Hillman
DMDaryush D. Mehta
Daryush D. Mehta