Course in Speech Recognition

Fall Semester 2003

Purpose

The purpose of this 5 p course is to give students with basic knowledge of speech technology a deeper understanding of techniques for speech recognition. The course includes practical assignments and a term paper on an individually selected topic.

The course consists of lectures, practical assignments and the writing of a term paper on an individually selected topic.

The following topics are treated in the course:

Introduction
Probability Theory
Pattern Recognition Techniques
Signal Processing and Feature Extraction,
Hidden Markov Models - training and decoding
Acoustic Modeling,
Adaptation to Speaker and Environment
Language Models
Basic and Large-vocabulary search strategies

Prerequisites

The course is aimed at students with a basic knowledge of speech technology (the equivalent of a GSLT level 1 course in Speech technology). Basic programming skills are useful as well as a knowledge of basic statistics and probability theory.

Schedule

Lectures

Date	Time	Room	Contents	Lecture Notes
September 19	10-12, 13-15, 15-16	Utställningen, TMH	1st Lecture	Slides_lecture_1
November 7	10-12, 13-15, 15-16	Utställningen, TMH	2nd Lecture	Slides_lecture_2 HTK Tutorial
November 28	10-12, 13-15	Utställningen, TMH	3rd Lecture	Slides_lecture_3_Ch_11 Slides_lecture_3_Ch_12
February 6	10-12, 13-15, 15-16	Utställningen, TMH	Closing seminar

Dead-lines

Date	Content
Nov 21	Select topic for term paper
December 5	Mail solutions to exercises to Mats
December 12	Mail selection of 2 papers for review
December 26	Author mail draft to reviewer
January 17	Mail review to author
January 24	Mail final paper to Mats

Reading material

The main text book for the course is Huang, Acero and Hon (2001), Spoken Language Processing.
A selection of papers will be used as additional reading material for topics not covered in the book.

Requirements

In order to pass the course the students must:

Complete the practical exercises
Prepare and present a term paper
Act as reviewer for two fellow student's term papers
Participate actively in the discussions in the closing seminar

Practical assignments

Exercises on speech recognition problems will be presented during the second lecture, Nov 7. The exercises will also be downloadable from Exercises

Send solutions not later than December 5 to Mats Blomberg matsb@speech.kth.se

Solutions will be presented during the closing seminar.

Term paper

During the course a term paper shall be prepared by each student and be presented during the closing seminar. The paper shall be reviewed by two fellow students.

Topic suggestions:
Choose a topic after discussion with the teacher (e.g. your own work and experiments)
Limitations in standard HMM and alternative approaches
Pronunciation variation and its importance for speech recognition
Language models for speech recognition
Search methods
Techniques for robust recognition of speech
Confidence measures in speech recognition
The role of prosody for speech recognition

Report your choice of topic and preliminary title by November 21 to Mats
Report your choice of 2 titles for review before Dec 12.

Chosen topics and assigned reviewers

Author	Title/Topic	Reviewer	Reviewer
Daniel Elenius	Adaptation techniques for children's speech recognition	Anna	Susanne
Genevieve Gorell	Language Modelling for Spoken Dialogue Systems; Grammar-Based and Robust Approaches Compared and Contrasted	Gustav	Magnus
Leif Grönqvist	Robust Methods for Automatic Transcription and Alignment of Speech Signals	Per-Anders	Daniel
Per-Anders Jande	Automatic Detailed Transcription of Speech Using Forced Alignment and Naïve Pronunciation Rules	Leif	Genevieve
Magnus Nordstrand	"The role of prosody for speech recognition"	Anna	Susanne
Anna Sandell	An exploration of hidden Markov models	Leif	Magnus
Susanne Schötz	Automatic prediction of speaker age using CART	Per-Anders	Gustav
Gustav Öquist	Speech Recognition for Context Awareness	Daniel	Genevieve

Closing seminar

The closing seminar includes:

Presentation of solutions to the practical assignments
Presentation and discussion of term papers

The closing seminar will take place at TMH, KTH 6 February 2004.

Seminar program

Time	Name	Title
10.15	Mats	Presentation of solutions to the exercises
11.00		Break
11.10	Anna	An exploration of hidden Markov models
11.35	Daniel	Adaptation techniques for children's speech recognition
12.00		Lunch
13.10	Genevieve	Language Modelling for Spoken Dialogue Systems; Grammar-Based and Robust Approaches Compared and Contrasted
13.35	Leif	Robust Methods for Automatic Transcription and Alignment of Speech Signals
14.00	Per-Anders	Automatic Detailed Transcription of Speech Using Forced Alignment and Naïve Pronunciation Rules
14.25	Susanne	Automatic prediction of speaker age using CART
14.50		Coffee break
15.10	Gustav	Speech Recognition for Context Awareness
15.35		Discussion
16.00		End

Teacher

Mats Blomberg matsb@speech.kth.se http://www.speech.kth.se/~matsb
Dept. Speech, Music and Hearing, KTH (Royal Institute of Technology)
SE-100 44 Stockholm, Sweden

How to get to TMH and some travel information

http://www.speech.kth.se/info/location.html