|Thesis topic proposal
Description of the research topic:
Articulatory-to-acoustic inversion deals with the problem of estimating
the acoustic signal based on
recordings of the position of articulatory organs. Many types of
equipments exist now to record the
articulatory movements, from simple video camera images that follow the
movement of the lips through
electromagnetic articulograpy (EMA) to special ultrasonic tools. The final
goal is to convert these
recordings into sound or a phonetic transcript with two main aims. First,
these tools might be a great
assistance for people who lost their ability to speak due to an operation
or accident. Second, this
approach might support speech recognition in extremely noisy environments.
These tools are called Silent
Speech Interfaces in general. The MTA-SZTE Research Group on Artificial
Intelligence has just started a
collaboration with the MTA-ELTE Lingual Articulation Research Group to
develop algorithms for silent
speech interfaces. The applicant should join this research, which applies
deep learning to the
articulatory-to-acoustic inversion task. The data sets are provided by the
Lingual Articulation Research
Group, while the deep learning tools and expertise comes from the Research
Group on Artificial
Intelligence. The task of the applicant is to get familiar with the
technologies used for silent speech
interfaces, invent and develop new algorithms, evalute the new ideas on
the data sets available in the
project, and finally to publish the new achievements.
1. T. Csapó, T. Grósz, G. Gosztolya, L. Tóth, A. Markó: DNN-based
Ultrasound-to-Speech Conversion for a Silent Speech Interface, Proc.
2. J. Freitas: An Introduction to Silent Speech Interfaces, Springer Verlag, 2017.
3. B. Denby, T. Schultz, K. Honda, T. Hueber, J. Gilbert, J. Brumberg:
Silent Speech Interfaces, Journal of Speech Communication, Vol. 52, No. 4, pp. 270-287, 2010.
Required language skills: english
Number of students who can be accepted: 1
Deadline for application: 2019-03-15