Maryam Najafian

Ph.D., MEng (First Class Honours)

Biographical Sketch

Maryam Najafian is a Research Scientist at the at the MIT IDSS, and previously a Postdoctoral Associate at the MIT CSAIL. She is an honorary Research Fellow at the University of Birmingham in England, school of Electrical, Electronic and Computer Engineering, where she got her PhD and Masters of Engineering degrees. Prior to joining MIT she served as a Research Scientist at the at the University of Texas at Dallas, Erik Jonsson School of Engineering and Computer Science.

She is currently working on two projects, namely "patent data analysis and citation network understanding", and "fairness, accountability, ethics and transparency in AI and Machine Learning ".
Prior to this, she was working on "dialect recognition and multi-dialect robust speech recognition", "child Autism detection through speaker diarization, location, and acoustic scene analysis", and "action recognition for patients with Aprexia or action disorganization syndrome".

She has published over 20+ refereed papers in leading journals and conferences. She is a reviewer for over 14 different IEEE, Springer, and Elsevier journals and conferences. Her in depth research on goodness of pronunciation scoring for children and her successful pitch for a child literacy support application has landed her the "PlanB Innovation Competition" and the "social enterprise UNLTD enterprise" awards in year on topic "Computerized English pronunciation learning". She also won the ﬁrst prize at the "IET enterprise" competitions on topic "Intelligent literacy system for children" and on topic "UK’s computerized child literacy support system". Later, she achieved the ﬁrst prize at the international 3minutes PhD thesis presentation (3MT) competition on topic "Accent robust speech recognition". She has been involved in implementation of a real-time human action recognition system in an award winning European funded project (Cogwatch) on cognitive rehabilitation of stroke patients with Aprexia or Action Disorganization Syndrome (AADS). In addition to research, Maryam has also been involved in student supervision, teaching, and organizing lab sessions for different projects and courses such as Speech, audio and music processing; Multimedia data; Optimization for economists; Maths for Applied computing; Statistics for economists; Digital systems and embedded computing, and Enterprise and entrepreneurship.

Research Interests

Maryam Najafian has 10+ years of leading and collaborating with multidisciplinary teams in AI. Her current (and past) research activities include applications of deep/recurrent/convolutional neural networks in solving a range of data science, and big data.
Her research interests are focused on but not limited to the following areas:

AI, machine learning, deep learning
Natural language processing (topic modeling, semantic analysis, and data mining)
Spoken language procesing (accent/noise robust speech recognition, dialect/language recognition, speaker identification, speaker diarization, goodness of pronunciation scoring)
Computer vision (image recognition, action recognition, object tracking)
Ethics, fairness, and transparency in AI

Current MSc/BSc students

Benjamin C Freed
Yazeed Awwad
Benny Siu Hon Ng

References and links to publications

Topic specific language modeling, feature level adaptation, and acoustic model adaption are applied to multi-genre multi-dialect speech recognition:

(1) M. Najafian, S. Khurana, S. Shon, A. Ali, J. Glass, "Exploiting Convolutional Neural Networks for Phonotactic Based Dialect Identification", in ICASSP, 2018. link
(2) M. Najaﬁan, W. Hsu, A. Ali, and J. Glass, "Automatic Speech Recognition Of Arabic Multi-genre Broadcast Media,” in ASRU, 2017. link
(3) M. Najaﬁan and M. J. Russell, “Improving automatic speech recognition on British English regional accents with limited resources,” Speech Communication,2018 (Under Revision). link
(4) M. Najaﬁan, S. Safavi, J. H. L. Hansen, and M. J. Russell “Improving speech recognition using limited accent diverse British English training data with deepneural networks,” in MLSP, 2016, pp. 1-8. link
(5) M. Najaﬁan, A. DeMarco, S. J. Cox, and M. J. Russell, “Unsupervised model selection for recognition of regional accented speech,” in INTERSPEECH, 2014, pp. 2967-2971. link
(6) M. Najaﬁan, S. Safavi, A. Hanani, and M. J. Russell, “Acoustic model selection using limited data for accent robust speech recognition,” in EUSIPCO,2014, pp. 1786-1790. link
(7) M. Najaﬁan and M. Russell, “Modelling accents for automatic speech recognition,”in EUSIPCO, 2015, pp. 1568. link
(8) M. Najaﬁan, S. Safavi, P. Weber, and M. J. Russell, “Identiﬁcation of British English regional accents using fusion of i-vector and multi-accent phonotactic systems,” in Ph. D. dissertation, University of Birmingham. link
(9) M. Najaﬁan, “Acoustic model selection for recognition of regional accented speech,” Ph. D. dissertation, University of Birmingham, 2016. link

Detection of Autism through location tracking, speaker diarization, and acoustic scene analysis applied to the child language environment:

(1) M. Najaﬁan, and J. H. L. Hansen,“Speaker independent diarization for child language environment analysis using Deep Neural Networks,” in SLT, 2016. link
(2) M. Najaﬁan, D. Irvin, Y. Luo, B. S. Rous, and J. H. L. Hansen, “Automatic measurement and analysis of the child verbal communication using classroom acoustics within a child care center,” in WOCCI, 2016. link
(3) M. Najaﬁan, D. Irvin, Y. Luo, B. S. Rous, and J. H. L. Hansen, “Employing speech and location information for automatic assessment of child language environments,” in SPLINE, 2016, pp. 65-69. link
(4) M. Najaﬁan, and J. H. L. Hansen, “Speaker diarization and speech activity detection fusion for detecting hot language areas during diﬀerent classroom activities using deep neural networks,” in ICASSP, 2017. link
(5) M. Najaﬁan, and J. H. L. Hansen, “Environment aware speaker diarization for moving targets using parallel DNN-based recognizers,” in ICASSP, 2017. link
(6) M. Najaﬁan, D. Irvin, B. S. Rous, and J. H. L. Hansen,“Automatic language environment monitoring using a child-adult turn-taking and a location tracking system,” in CSL, 2018 (Under Revision). link

Natural language processing, speaker recognition and accent recognition systems for high-confidence biometric presentation using phonotactic, lexical and acoustic features:

(1) S. Khurana, M. Najafian, Ahmed Ali, Tuka Al Hanai, Y. Belinkov, and J. Glass,“QMDIS: QCRI-MIT Advanced Dialect Identification System,” in Interspeech, 2017. link
(2) M. Najaﬁan, M. J. Russell, and S. D’Arcy “Identication and visualisation of accented speech using i-vector and phonotactic methods and their fusion,” in CSL 2017 (Under Revision).link
(3) M. Najaﬁan, S. Safavi, P. Weber, and M. J. Russell, “Identiﬁcation of British English regional accents using fusion of i-vector and multi-accent phonotactic systems,” in ODYSSEY, 2016, pp. 132-139.link
(4) S. Safavi, M. Najaﬁan, A. Hanani, M. J. Russell, P. Jancovic and M. J.Carey “Speaker recognition for children’s speech,” INTERSPEECH, 2012, pp.1836-1839.link
(5) S. Safavi, M. Najaﬁan, A. Hanani, M. Russell and P. Jancovic, “Comparison of speaker veriﬁcation performance for adult and child speech,” WOCCI, 2014,pp. 27-31 . link

Image and sensor data processing for action recognition systems that help stroke patients with Aprexia or Action Disorganization Syndrome (AADS):

(1) R. Nabiei, M. Najaﬁan, M. Parekh, P. Jancovic, and M. Russell, “Delay reduction in real-time recognition of human activity for stroke rehabilitation,” in SPLINE, pp. 70-74, 2016. link
(2) R. Nabiei, M. Najaﬁan, P. Jancovic, and M. J. Russell “Rehabilitation Of stroke patients with parallel GMM- and DNN-HMM based human activity recognition using instrumented objects,” in CHI, 2018 (Under Revision). link
(3) M. Liu, H. Liu, C. Chen and M. Najaﬁan, ”Energy-based global ternary image for action recognition using sole depth sequences,” 3D Vision, 2016 link
(4) H. Su, S. Tian, Y. Cai, Y. Sheng, C. Chen and M. Najaﬁan, “Optimized extreme learning machine for urban land cover classiﬁcation using hyperspectral imagery” in Frontiers of Earth Science 2016, Springer, pp. 1-8. link
(5) C. Chen and M. Najaﬁan, ”IPJ: Informative Pairwise Joints For 3D human action recognition,”in WACV, 2016. link

Links to my talks

Speaker independent diarization for child language environment analysis using Deep Neural Networks and i-Vectors. link
Improving speech recognition using limited accent diverse British English training data with acoustic model and data selection. link
Rehabilitation of stroke patients with aprallel DNN based human activity detectors using instrumented objects. link