Kartik Audhkhasi's Homepage

Ph.D. Student (since Fall 2008)
Signal Analysis and Interpretation Lab (SAIL),
Advisor: Prof. Shrikanth Narayanan,
Electrical Engineering Department,
Viterbi School of Engineering,
University of Southern California, Los Angeles, U.S.A.

Previous Affiliations:
2003-2008: BTech. in Electrical Engineering,
MTech. in Information and Communication Technology,
Indian Institute of Technology - Delhi, India

Research Interests

• Specific: Statistical modeling, learning and inference in machine learning problems involving multiple perspectives. For example - noisy labels from multiple experts, ensemble methods in machine learning, crowd-sourcing. Applications in speech processing and recognition, behavioral and multi-modal signal processing.
• General: Machine learning, graphical models, speech recognition, analytics and processing.

Publications

1. Kartik Audhkhasi, Panayiotis Georgiou and Shrikanth Narayanan, "Analyzing quality of crowd-sourced speech transcriptions of noisy audio for acoustic model adaptation", Proc. ICASSP, 2012, Kyoto, Japan. (Preprint)
2. Kartik Audhkhasi, Abhinav Sethy, Bhuvana Ramabhadran, and Shrikanth Narayanan, "Creating ensemble of diverse maximum entropy models", Proc. ICASSP, 2012, Kyoto, Japan. (Preprint)
3. Kartik Audhkhasi, Panayiotis Georgiou and Shrikanth Narayanan, "Reliability-weighted acoustic model adaptation using crowd sourced transcripts", Proc. InterSpeech 2011, Florence, Italy. (Preprint)
4. Kartik Audhkhasi and Shrikanth Narayanan, "Emotion classification from speech using evaluator reliability-weighted combination of ranked lists", Proc. ICASSP 2011, Prague, Czech Republic. (Preprint)
5. Kartik Audhkhasi, Panayiotis Georgiou and Shrikanth Narayanan, "Accurate transcription of broadcast news speech using multiple noisy transcribers and unsupervised reliability metrics", Proc. ICASSP 2011, Prague, Czech Republic. (Preprint)
6. Kartik Audhkhasi and Shrikanth Narayanan, "Data-dependent evaluator modeling and its application to emotional valence classification from speech", Proc. InterSpeech 2010, Makuhari, Japan. (Preprint)
7. Qun Feng Tan, Kartik Audhkhasi, Panayiotis Georgiou, Emil Ettelaie and Shrikanth Narayanan, "Automatic speech recognition system channel modeling", Proc. InterSpeech 2010, Makuhari, Japan. (Preprint)
8. Kartik Audhkhasi, Panayiotis Georgiou and Shrikanth Narayanan, "Lattice-based lexical cues for word fragment detection in conversational speech", Proc. ASRU, Merano, Italy, pp. 568-573, December 2009. (Preprint)
9. Kartik Audhkhasi, Kundan Kandhway, Om Deshmukh and Ashish Verma, "Formant-based technique for automatic filled-pause detection in spontaneous spoken English", Proc. ICASSP, Taipei, Taiwan, pp. 4857-4860, April 2009. (Preprint)
10. Om Deshmukh, Kundan Kandhway, Ashish Verma and Kartik Audhkhasi, "Automatic evaluation of spoken English fluency", Proc. ICASSP ,Taipei, Taiwan, pp. 4829-4832, April 2009. (Preprint)
11. Kartik Audhkhasi and Ashish Verma, "Keyword search using modified minimum edit distance measure", Proc. ICASSP, Honolulu, Hawaii, pp. 929-932, April 2007. (Preprint)
12. Kartik Audhkhasi and Arun Kumar, "Two-scale auditory feature based non-intrusive speech quality evaluation", IETE Journal of Research, vol. 56, no. 2, pp. 111-118, March-April 2010. (Web link)
13. Kartik Audhkhasi, "Automatic evaluation of fluency in spoken language" (Review article), IETE Technical Review, vol. 26, no. 2, pp. 108-114, March-April 2009. (Web link) (IETE M. N. Saha Memorial Award for Best Application Oriented Paper, 2010)

Patents

1. Kartik Audhkhasi, Om D. Deshmukh, Kundan Kandhway and Ashish Verma, "Automatic evaluation of spoken fluency", US Patent Appl. No. 12/541,927, filed August 12, 2009.

Curriculum Vitae: PDF

Projects at SAIL

• An Integrated Approach to Creating Enriched Speech Translation Systems (NSF)
  Currently involved in developing techniques for using crowd-sourced data in training conversational ASR systems in Mexican Spanish and English. Previously involved in developing features and algorithms for detecting anomalous events from spontaneous speech. Worked on detection of filled pauses, word fragments and repetitions using acoustic-prosodic and lexical features.
• ASR Training (various funding agencies)
  Involved in training/testing acoustic and language models for speech recognition in diverse languages.
• AM Adaptation in ASR
  Involved in implementing maximum likelihood linear regression (MLLR) for adapting acoustic models in a large vocabulary ASR system.

Dual Degree Project at IIT-Delhi

Non-Intrusive Speech Quality Evaluation (Advisor: Dr. Arun Kumar, Center for Applied Research in Electronics, IIT-Delhi)

SAIL Machine Learning Reading Group and Lecture Series: Webpage

Teaching Assistance

Why is teaching important for researchers?
*Industry folks: I do not necessarily agree with the last sentence on the above link ;)

• At USC
  Fall 2011 - EE 562a: Random Processes in Engineering (Prof. Robert Scholtz)
  
  • Discussion session - 5:00 to 5:50 pm, Friday, OHE 100C
    
  • Office hours - 3:00 to 6:00 pm, Thursday, PHE 320
    
  • TA Evaluation Summary*
    
  
  Spring 2011 - EE 562a: Random Processes in Engineering (Prof. Robert Scholtz)
  
  • Discussion session - 8:30 to 9:20 am, Monday, OHE 100B
    
  • Office hours - 9:30 to 11:30 am, Monday, PHE 320
  • TA Evaluation Summary*
    
  
  Fall 2010 - EE 562a: Random Processes in Engineering (Prof. Robert Scholtz)
  
  • Discussion session - 5:00 to 5:50 pm, Friday, OHE 100C
    
  • Office hours - 3 to 5 pm, Tuesday, PHE 320
    
  • TA Evaluation Summary*
    
  *Ratings on a 1-5 Likert scale
  EE 562a Syllabus: Click here
  
  
• At IIT-Delhi
  Fall 2007 - EEP 306: Communication Engineering Laboratory (Prof. Ranjan Mallik)
  Spring 2008 - EEL 316: Digital Communications (Prof. Ranjan Mallik) (laboratory sessions)

Industrial Experience

• IBM T. J. Watson Research Center, Yorktown Heights, New York (Human Language Technologies Group, May-August 2011)
  Mentors: Abhinav Sethy, Bhuvana Ramabhadran
  Worked on training diverse ensemble of maximum entropy classifiers for natural language processing.
• IBM India Research Lab, New Delhi (Human Language Technologies Group, May-July 2008)
  Mentors: Om Deshmukh, Ashish Verma
  Worked on filled pause detection and automatic fluency evaluation for spoken English.
• IBM India Research Lab, New Delhi (Human Language Technologies Group, May-July 2006)
  Mentor: Ashish Verma
  Worked on keyword spotting from spoken English.

Select Graduate Coursework

• At USC
  Neural and Fuzzy Systems (in progress), Optimization: Theory and Algorithms, Machine Learning, Advanced Artificial Intelligence (Graphical Models), Empirical Methods in Natural Language Processing, Mathematical Pattern Recognition, Statistics for Engineers, Random Processes in Engineering, Wavelets.
• At IIT-Delhi
  Human and Machine Speech Communication, Detection and Estimation Theory, Statistical Signal Processing, Multimedia Systems, Signal Theory, Computer Vision, Image Processing, Selected Topics in Communications (MIMO Communications), Neural Networks and Machine Learning.

Other Interests

Body-building, Racquetball, Dux Ryu Ninjistu, Cricket

Contact Information

Office: EEB 405, 3740 McClintock Avenue,
University of Southern California,
Los Angeles, CA 90089-2564
Office phone: (213)-740-4660
E-mail: