Exploring infrastructure for Dutch speech recognition
Due to developments in AI, the world of automatic speech recognition (ASR) is rapidly changing. New ASR systems seem to provide overwhelmingly accurate transcription of speech. But how do these systems perform under atypical conditions and in large scale applications?
- 25 Jun 2024
-
Time13:00-17:00h.
-
LocationSURF Utrecht
ASR systems that have become available on the market recently such as Whisper, seem to provide overwhelmingly accurate transcription of speech. But how do these systems perform under atypical conditions? For example, in the case of dialects, children or elderly speech or speech from non-native Dutch speakers? What happens if there are multiple speakers, cross talk and background noises? And, what to do if you want to transcribe very large amounts of speech data? What's the best way to handle this in a more (infra)structural way?
In this seminar, we will show examples from different application areas and discuss practical, operational, and strategic aspects of:
- The necessity of making highly quality (Dutch) speech recognition engines available in research and educational context, e.g., to transcribe speech from lectures, interviews or meetings to text.
- Whether it is important or not that speech recognition engines can be upgraded when better engines or models become available, or that different versions of models can be selected for specific tasks (e.g., specific types of speech).
- How open standards can be applied and ‘explainability’ can be fostered (how are models created, using which data sets, performance specifications) as much as possible.
This seminar has not the aim to facilitate research on speech technology, rather it addresses the use of existing speech technology solutions, and how this can be (further) optimized: let's move together towards sustainable solutions for research and education!
Program
Chair: Annette Langedijk (SURF)
12:45h | Doors open |
13:00h | Welcome and Setting the Stage (Roeland Ordelman, CTO CLARIAH) |
13:10h |
User perspectives on ASR
|
14:30h | Break |
15:00h |
Technological perspectives on ASR
|
15:50h | Panel discussion |
16:30h | Closing and networking reception |
Note: The language of the meeting will be English.
For whom?
The event is of particular interest to:
- Researchers, educators and support staff from various disciplines interested in the application of automatic speech recognition
- Research & education infrastructure providers
Location
SURF Utrecht (Kantoren Hoog Overborch - Hoog Catharijne)
Moreelsepark 48
3511 EP Utrecht
Route description
Organisation
SURF in cooperation with Stichting Open Spraaktechnologie.