Open seminar on speech processing
This open seminar will explore the state of the art in speech recognition algorithms.
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
- Janiszewskiego 7/9
- Wrocław, Dolnoslaskie
- Poland 50-372
- Building: C5
- Room Number: 409
- Click here for Map
- Contact Event Host
- Co-sponsored by Wroclaw University of Science and Technology, Faculty of Electronics, Photonics and Microsystems
Speakers
Ryszard Makowski of Wroclaw University of Science and Technology
Automatic words recognition
This paper focuses on speech signal preprocessing, its parameterization, and methods for building acoustic models for automatic speech recognition (ASR) systems. The presented methodology for solving the recognition problem is based on a hybrid approach, utilizing probabilities based on GMM distributions with diagonal and full covariance matrices, and modeling based on statistical n-gram models (bigrams and trigrams) in slightly modified hidden Markov series.
Particular emphasis is placed on the initial signal acquisition and parameterization stages, as well as on building acoustic state models, what is resulting from the author's main research specialization, i.e. digital random signal processing. The presentation details the implemented methodology and presents and analyzes the obtained results on the recognition performance of word sequences.
Email:
Ryszard Makowski of Wroclaw University of Science and Technology
Automatic words recognition
Email: