Automatic Speech Recognition - Advanced Topics (2008)

Course responsibles:

Coordinator: Torbjørn Svendsen

Lecturers: Torbjørn Svendsen, Magne H. Johnsen, Marco Siniscalchi, Ingunn Amdal, Jacques Koreman, Wim van Dommelen, Odette Scharenborg

Dates and location:

The NGSLT course on Advanced Topics in Automatic Speech Recognition will be held at the Norwegian University of Science and Technology (NTNU) in Trondheim, Norway. The course will be held in two parts, the first on October 22-24, 2008 and the second half on November 24-26, 2008.

Course overview:

The course aims to provide the necessary background for understanding recent approaches to automatic speech recognition. As it is not expected that all the students will have an intimate knowledge of speech recognition, the course will include traditional approaches to speech analysis and recognition as well as knowledge driven and detector based approaches. The course will give 7.5 ECTS credits.

Topics covered include:

• Phonetics (basic phonetics, articulatory phonetics, distinctive phonetic features)
• Speech signal processing (basics, short time spectral estimation, models for speech analysis, time and frequency domain techniques for phonetic feature detection)
• Statistical speech recognition (basic theory, HMM, ANN, Graphical models, Units for ASR, Acoustic modeling, Lexica, Language modeling)
• Decoding (Viterbi decoding, Finite state machines)
• Performance evaluation.

Detailed course content and schedule

Literature for the course

Lecture notes and presentations

Prerequisites:

The course is intended for PhD students with a basic knowledge of speech technology.

Term paper:

The NGSLT students will be required to write a term paper related to the topics covered by the course. Each student is required to propose a topic, no later than December 1, 2008. The report is to be submitted by January 15, 2009.

The term paper can be a literature study, or based on experimental work.

Supervision for the term paper will be organized by Torbjørn Svendsen. If your local supervisor is willing to act as supervisor, that is fine, but we will need to be notified. Otherwise, one of the lecturers involved in the course will act as your supervisor. Proposals for topics and supervisors are to be sent to Torbjørn.

Miscellaneous information:

The course participants will include PhD students at NTNU and the course will be held regardless of the number of NGSLT students.

Course times: The first part of the course will start at 9:15 on Oct 22 and end at 16:00 on Oct 24. The second part will start at 9:15 on Nov 24 and end at 16:00 on Nov 26.

Course location: Room B-343, EE block B (O.S.Bragstads pl 2B), NTNU Gløshaugen, Trondheim (see Campus map)