Module 39-M-Inf-SSV Speech Signal Processing

Attention: This page shows a discontinued module offer.

Faculty

Person responsible for module

Regular cycle (beginning)

Discontinued

Credit points and duration

10 Credit points

For information on the duration of the modul, refer to the courses of study in which the module is used.

Competencies

Non-official translation of the module descriptions. Only the German version is legally binding.

In this lecture students will learn basic methods of automatic speech recognition (ASR) as they are used both in research prototypes as well as in off-the-shelf ASR systems. After this lecture students should be able to critically examine the performance of ASR systems and to train and evaluate an ASR system such as Esmeralda or HTK.

Content of teaching

The lecture Speech Recognition focuses on methods to automatically derive an orthographic representation from spoken speech as used in automatic dictating machines or for the speech-based control of technical systems. We will start-off with an overview of the human articulatory system in order to better understand the components of the speech signal in terms of the source-filter model as well as phonetic phenomena such as coarticulation or reduction. This forms the basis for the signal processing methods where the different components of source and filter are decomposed. One important part of the lecture are Hidden-Markov-Models (HMMs) which still represent the paradigm of state-of-the-art ASR approaches. The mathematical basis of this statistical modeling approach will be discussed and algorithms for training and decoding of HMMs given a speech signal and an annotation for training will be presented in detail. Several variants of ASR systems will be discussed.

The lecture "Application oriented speech processing" will present concrete implementations of the algorithms discussed in the lecture "Speech Recognition". Within the exercises advanced ASR techniques will be derived at a theoretical level and will be implemented and evaluated in group projects.

Alternatively to this lecture selected topics of ASR will be discusse in a seminar. In the seminar participants are expected to prepare and give a presentation on a specific topic as well as provide a written summary.

Recommended previous knowledge

Recommended Competences: Competences as they can be achieved with the Module 39-Inf-MK Pattern Recognition

Necessary requirements

Explanation regarding the elements of the module

Notes on course selection:
You must attend either the lecture and tutorial Application oriented speech processing or the seminar speech processing.

Ungraded / Graded Module Examination:
The (partial) examination of the module can be performed as "ungraded" in some study programs at the students choice. Before the examination a respective determination must be carried out, a later modification (graded - ungraded) is impossible. If the "ungraded" option is chosen, it is not possible to include this module in a study program where this module is deemed to enter the calculation of the overall grade.

Module structure: 1 SL, 0-1 bPr, 0-1 uPr 1

Courses

Anwendungsorientierte Sprachverarbeitung
Type lecture
Regular cycle SoSe
Workload5 30 h (15 + 15)
LP 1

Alternatively to lecture and exercises a seminar language processing can be chosen.

Anwendungsorientierte Sprachverarbeitung
Type exercise
Regular cycle SoSe
Workload5 90 h (45 + 45)
LP 3 [SL]

Alternatively to lecture and exercises a seminar language processing can be chosen.

Speech Recognition
Type exercise
Regular cycle WiSe
Workload5 30 h (15 + 15)
LP 1
Speech Recognition
Type lecture
Regular cycle WiSe
Workload5 90 h (45 + 45)
Sprachverarbeitung
Type seminar
Regular cycle SoSe
Workload5 120 h (30 + 90)
LP 4 [SL]

Alternatively to lecture and exercises a seminar language processing can be chosen.


Study requirements

Allocated examiner Workload LP2
Teaching staff of the course Anwendungsorientierte Sprachverarbeitung (exercise)

Oral presentation (15-25 min.)

see above see above
Teaching staff of the course Sprachverarbeitung (seminar)

Oral presentation (15-25 min.)

see above see above

Examinations

oral examination
Allocated examiner Teaching staff of the course Speech Recognition (lecture)
Weighting without grades
Workload 60h
LP2 2

In some degree programmes of the Faculty of Technology, the module examination can also be "ungraded" at the student's discretion (see explanations of the module elements and the respective subject-specific regulations). If the ungraded option is selected, it is not possible to use this module for a degree programme in which this module is taken into account in the overall grade calculation.
See below for explanations of this examination (graded examination option).

oral examination
Allocated examiner Teaching staff of the course Speech Recognition (lecture)
Weighting 1
Workload 60h
LP2 2

Oral examination (15-25 min.) about the contents of lecture and exercises.

Further notices

Bei diesem Modul handelt es sich um ein eingestelltes Angebot. Ein entsprechendes Angebot, um dieses Modul abzuschließen, wurde bis maximal Sommersemester 2019 vorgehalten.
Bisheriger Angebotsturnus war jedes Wintersemester.

The module is used in these degree programmes:

Degree programme Profile Recom­mended start 3 Duration Manda­tory option 4
Data Science / Master of Science [FsB vom 06.04.2018 mit Änderungen vom 01.07.2019, 02.03.2020, 21.03.2023 und 10.12.2024] Variante 1 3. two semesters Compul­sory optional subject
Data Science / Master of Science [FsB vom 06.04.2018 mit Änderungen vom 01.07.2019, 02.03.2020, 21.03.2023 und 10.12.2024] Variante 2 3. two semesters Compul­sory optional subject
Intelligent Systems / Master of Science [FsB vom 27.07.2018 mit Änderung vom 04.06.2020] 1. two semesters Compul­sory optional subject
Intelligent Systems / Master of Science [FsB vom 17.12.2012 mit Änderungen vom 15.04.2013, 01.04.2014, 15.10.2014, 02.03.2015 und Berichtigung vom 17.11.2014] 1. two semesters Compul­sory optional subject
Informatics for the Natural Sciences / Master of Science [FsB vom 30.09.2016 mit Berichtigung vom 10.01.2017 und Änderungen vom 15.09.2017, 02.05.2018, 04.06.2020 und 31.03.2023] 1. o. 3. two semesters Compul­sory optional subject
Informatics for the Natural Sciences / Master of Science [FsB vom 17.12.2012 mit Änderungen vom 15.04.2013, 01.04.2014, 15.10.2014, 02.03.2015, 01.12.2015 und Berichtigungen vom 01.04.2014, 17.11.2014 und 12.07.2017] 1. o. 3. two semesters Compul­sory optional subject

Automatic check for completeness

The system can perform an automatic check for completeness for this module.


Legend

1
The module structure displays the required number of study requirements and examinations.
2
LP is the short form for credit points.
3
The figures in this column are the specialist semesters in which it is recommended to start the module. Depending on the individual study schedule, entirely different courses of study are possible and advisable.
4
Explanations on mandatory option: "Obligation" means: This module is mandatory for the course of the studies; "Optional obligation" means: This module belongs to a number of modules available for selection under certain circumstances. This is more precisely regulated by the "Subject-related regulations" (see navigation).
5
Workload (contact time + self-study)
SoSe
Summer semester
WiSe
Winter semester
SL
Study requirement
Pr
Examination
bPr
Number of examinations with grades
uPr
Number of examinations without grades
This academic achievement can be reported and recognised.