The conference dinner will be held at Simon's restaurant at Groot Constantia.

Delegates attending the dinner are required to select one item from each course in the restaurant menu (available here) at least one week in advance.  Please email your selections to Karin van Wyk. 

Thursday 27 November

David van Leeuwen:  The difference that South Africa has made to Speaker Recognition

 Automatic speaker recognition is an area of speech technology that has
received much attention from speech researchers in recent years.  Some
believe that it is the cleanest of all speech related recognition
problems.  Although simple in its formulation, the speaker recognition
problem appears to have an intricate relation with its application.
Text independent Speaker Recognition can be seen as a pattern
recognition problem, where features are highly variable sequences
related to a single source.  The task is to detect whether the source
is of known identity.

The engineering of speaker recognition systems depends largely on the
availability of example material, in this case speech recordings of
thousands of different speakers.  Performance is driven by
international benchmark evaluations which have been carried out almost
every year since 1996 by the National Institute of Standards and
Technology in the United States.  These competitive evaluations
donate new evaluation data to the research community, which guides
research directions.  In recent years, the challenge of channel and
session variability has been the focus of these evaluations.

In this presentation, the typical characteristics of the speaker
recognition approach are reviewed, and an overview of the machine
learning techniques employed is given.

In recent years, new approaches to the presentation of the speaker
recognition output have been developed.  This way of presentation makes
the technology applicable in a wider range of applications without the
need of recalibration.  Both in the attempts to overcome channel
variability and the application-independent presentation of speaker
recognition output researchers from South Africa have played an
important role.

Friday 28 November

Thomas Hain: The Careful Listener: Speech Processing in Meetings

 Meetings form an essential part of life for many people and the time spent in face to face meetings is ever increasing while more and more people complain about inefficiency, lack of planning and loss of information. Meetings have to be postponed due to lack of information at the time, essential participants that could not attend or deviation from the real topics at hand. While we are normally very eager to use tools that help to increase productivity in many areas, meetings seem to have been mostly excluded in this quest.

Under the AMI and AMIDA projects observant technologies are developed that aim to assist humans in their tasks rather than replacing their functions. While many of these technologies use several modalities (such as video, speech, handwriting, etc) at the same time, the most important information to date can be derived from speech signals alone. However, most known algorithms have to be altered to cope with the complex acoustic situation and special information not relevant in other domains can be derived.

In this presentation a brief overview of the AMIDA project is given, followed by a discussion of required information for several applications. The information related to speech signals are the speakers identity and location, the timing, the content, the presentation style. Hence speaker diarisation, speaker tracking, and speech recognition are at the core of speech technologies used. The presentation will give an overview of state of the art systems for meetings and their performance. Since processing should be minimally invasive microphone array processing is fundamental to all systems presented. Examples of systems for higher level information extraction using the output of these speech processing algorithms are given.
 

Papers accepted for oral presentations

These papers will appear in full in the conference proceedings.  Any revisions (optional or required) to papers must be received by the end of Wednesday 12 November.  Each oral will be allocated 20 minutes, including questions.

Papers accepted for work-in-progress poster presentations

Abstracts of these papers will appear in the conference proceedings, and an electronic document containing these papers will be made available.  Posters should not exceed A0 in size, but the format is otherwise unspecified.