Speech Reading Club

Aims
Structure
Themes
Assessment
Themes and Topics References

If there is a theme, or topic within a theme, that you are very interested in that is not listed contacted me. We can discuss including it.

Aims

Structure

• By 18th November 2004 students will be asked to make a decision whether they are taking the Speech Reading Club module. If the module is chosen a preference for a particular theme/topic may be expressed. The maximum number of students allowed to take the Speech Reading Club is 60% of the CSTIT course.
  
  
• Students will be divided into groups of 3-5. Associated with each group there is a particular theme. Each student in the group will also be assigned a particular topic within that theme. Where possible the assignment of themes and topics will reflect any preferences expressed. However, not all themes or topics will be run.
  
  
• Papers associated with each theme/topic will be distributed, or made available on the web, at the start of Lent term. For each theme a 1 hour discussion session with the theme organiser will be organised during weeks 3 and 4 of Lent term. Specific problems may also be discussed via email.
  
  
• During weeks 5 to 8 of Lent term a one to two hour presentations of each theme will be given. These will consist of a 5-10 minute overview of the theme, jointly prepared by all members of the group. Each student will then give a 15 minute presentation of their particular topic. Each presentation is followed by a short discussion about topic lead by the student and theme organiser.
  
  ALL students taking the Speech Reading Club module must attend ALL presentations.
  
  
• Each student writes an essay that describes the general issues in the their theme area and a detailed discussion about their particular topic. The maximum word length for the essays is 5000 words. The essay must be handed in by 26th April 2005.

• 18th November 2004: decision on whether to do the speech reading club module.
• 25th November 2004: list of topic and theme assignments circulated.
• Weeks 5-8 Lent term: speech reading club presentations.
• 26th April 2005: hand-in date for the speech reading club essays.

Themes

• Speech Synthesis - theme organiser Paul Taylor
• Discriminative Training - theme organisers Mark Gales and Bill Byrne
• Speaker Adaptation - theme organiser Mark Gales
• Noise Robustness - theme organiser Mark Gales
• Statistical Machine Translation - theme organiser Bill Byrne
• Bayesian Networks and Segment Models - theme organiser Mark Gales

Assessment

Themes and Topics References

• X. Huang, A. Acero and H-W Hon, Spoken Language Processing, Prentice Hall.

• Speech Synthesis
• Discriminative Training
• Speaker Adaptation
• Noise Robustness
• Statistical Machine Translation

• Acoustics of Speech Production
• basics of how sound wvaes are produced and how they travel
• source/filter model of speech production
• models of vocal tracts as tubes
• sound waves in tubes
• sound sources, formants, linear prediction
• Prosodic Modelling
• basic prosody models
• automatic recognition of prosody
• synthesis of prosody
• using prosodic information in speech recogntion

• Tobi home page
• Taylor, Paul (book chapter), Acoustic Theory of Speech Production
• Black AW, Hunt A Generating f0 contours from ToBI labels using linear regression
• P. Taylor. Analysis and synthesis of intonation using the tilt model. Journal of the Acoustical Society of America, 107(3):1697-1714, 2000.
• Ann Syrdal, Gregor Moehler, Kurt Dusterhoff, Alistair Conkie, and Alan Intonation modeling. In 3rd ESCA Workshop on Speech Synthesis, pages 305-310, Jenolan Caves, 1998.
• Comparing CART and Fujisaki Intonation Models for Synthesis
• A Method for Automatic Extraction of Fujisaki-Model Parameters
• Elizabeth Shriberg et al. Can prosody aid the automatic classification of dialog acts in conversational speech? Language and Speech, 41(3-4), 1998. R.
• MIT lecture notes
• Huang et al. Spoken Language Processing, Chapter 6
• D. Ellis (lecture notes)
• P. Taylor Acoustic Theory of Speech Production (book chapter)
• P. Taylor Introduction to Signals and Filters (book chapter)

• Maximum mutual information (MMI) and frame discrimination (FD) training criteria;
• Minimum classification error rate training criterion;
• Discriminative training for large vocabulary systems;
• Discriminative methods for speaker adaptation and feature extraction.

• (*) M. Gales (2000) lecture note extract
• P.C. Woodland and D. Povey (2002) Large Scale Discriminative Training of Hideen Markov Models for Speech Recognition Computer Speech and Language, Vol. 16, pp. 25-47.
• R. Schluter (2000). Investigations on Discriminative Training PhD thesis (500Kb Compressed).
• L.F. Uebel and P.C. Woodland (2001). Discriminative Linear Transforms for Speaker Adaptation Adaptation Methods for Speech Recognition, ISCA Tutorial and Research Workshop.
• D. Povey and P.C. Woodland (2002) Minimum Phone Error and I-Smoothing for Improved Discriminative Training ICASSP.

• Speaker clustering, eigenvoices and cluster adaptive training;
• Linear model-based adaptation schemes and adaptive training, maximum likelihood linear regression (MLLR) and constrained MLLR;
• Maximum a-posteriori (MAP) adaptation schemes and extensions, e.g. adaptation by correlation.

• (*) M. Gales (2000) lecture notes
• C-H. Lee and Q. Huo (2000). On Adaptive Decision Rules and Decision Parameter Adaptation for Automatic Speech Recognition, Proceedings of IEEE Vol. 88.2.
• P.C. Woodland (2001) Speaker Adaptation for Continuous Density HMMs: A Review ITWR Adaptation Workshop.

• Enhancement schemes (including model-based enhancement);
• Predictive and adaptive model-based compensation schemes;
• Inherently robust frontends and models.

• (*) M. Gales (2000) lecture notes
• J. Ming and F.J. Smith (2001). Union: a model for partial temporal corruption of speech Computer Speech and Language, Vol. 15, pp. 217-231.
• M.J.F. Gales (1998). Predictive Model-Based Compensation Schemes for Robust Speech Recognition. Speech Communication Volume 25 1998.
• H. Hermansky (1997). Should Recognisers Have Ears?. Proceedings ESCA Workshop.
• H. Liao and M.J.F. Gales (2004) Uncertainty Decoding for Noise Robust Automatic Speech Recogntion.
• M.J.F. Gales (2001) Adaptive Training for Robust ASR, ASRU 2001.

• Bitext Alignment
• Word Alignment in Bitext
• Models and Algorithms for Statistical Translation

• W. Gale and K. Church A Program for Aligning Sentences in Bilingual Corpora Association for Computational Linguistics. 1991
• R. Moore Fast and Accurate Sentence Alignment of Bilingual Corpora Proceedings, 5th Conference of the Association for Machine Translation in the Americas, Tiburon, California, Springer-Verlag, Heidelberg, Germany, pp. 135-244.
• P. Brown and S. Della Pietra and V. Della Pietra and R. Mercer The Mathematics of Statistical Machine Translation: Parameter Estimation Association for Computational Linguistics. 1993
• F. Och and H. Ney A Systematic Comparison of Various Statistical Alignment Models Association for Computational Linguistics, 2003
• K. Knight and Y. Al-Onaizan Translation with Finite-State Devices Proceedings of the 4th AMTA Conference, 1998.
• F. Och and H. Ney The Alignment Template Approach to Statistical Machine Translation Computational Linguistics, December 2004, pp 417--450
• S. Kumar, Y. Deng, and W. Byrne A weighted finite state transducer translation template model for statistical machine translation Journal of Natural Language Engineering. To appear.

• Dynamic Bayesian networks and graphical models for speech recognition;
• Distributed representations for speech recognition;
• Linear dynamical and factor analysed systems;
• Efficient covariance modelling.

• (*) Kevin Murphy (2001) A brief Introduction to Graphical Models and Bayesian Networks
• S Roweis and Z. Ghahramani (1999) A Unifying Review of Linear Gaussian Models Neural Computation, Volume 11, Pages 305-345.
• A-V I Rosti and M.J.F. Gales (2001) Generalised Linear Gaussian Models Technical Report Cambridge University.
• M. Ostendorf (1996) From HMMs to Segment Models: A Unified Review of Stochastic Modeling for Speech Recognition IEEE Transactions on Speech and Audio Processing.
• M.J.F. Gales (2002) Transformation Streams and the HMM Error Model Computer Speech and Language, Volume 16, Pages 225-243.