Ondřej Plátek

Main Research Interests


Applications in Dialogue Systems, Speech Recogniton and Information Retrieval


  • (Inverse) Reinforcement Learning and (Recurrent) Neural Networks are fascinating tools
  • Open-source, Data-Driven, Reproducible, Linux, Clusters of machines (SGE/Spark/Hadoop)
  • Language preferences in following order: Python, C#, F#, C++, (Java, Scala)


Curriculum Vitae

  • since 2014: PhD student at the Faculty of Mathematics and Physics, Charles University in Prague
  • 2014: Mgr. in Theoretical Computer Science, Charles University in Prague
    Master's Thesis on Speech recognition using Kaldi (pdf)

For further details see https://www.linkedin.com/in/ondrejplatek


NPRG045NSWI095 labs


 Suggestions for projects/Bachelor/Master thesis

keywords: ročníková práce, bakalářská práce, magisterská práce, dobrovolná pomoc :)

What can I offer you as a supervisor?

  1. Your project will be used in our SDS group.
  2. Your project will be on the edge of industry and research.
  3. Progress meetings/consulting every week.
  4. Contacts in academia and industry not only in ČR.

Contact me, if you have your own topic and you are looking for supervisor! ...  However, check that you can answer "yes" to questions below:

  1. Can you explain me in five sentences how your topic relates to following areas:
    • dialogue systems,
    • speech recognition,
    • voice command control of robots,
    • text to speech?
    • If you are very confident apply with any application which will help disabled people.
  2. Are you willing to work on your project 16 hours a week for one year?
    • I hope you will manage to be faster, but ask some of your older collegues how long one needs for a thesis.
  3. Will you publish your code under Apache 2.0 license?
  4. Will you implement your code in  Python, C#, C++, F#, Java or Scala and deploy it on Ubuntu/Debian distribution?



Dialogue Gamification  [e.g. Run Neo Run] 

  • Required: Finish implementation of simple game. Collect dialogues
  • We can offer resources for running/advertising/crowdsourcing the game.
  • Required platform: HTML5, python backend
  • Collecting dialogues through simple but attractive game: E.g. see https://github.com/oplatek/run-neo-run (may be used as a baseline)
  • Convenient for any type of project. Complexity is in level of details and scalability.

Voice Control for Robots

  • Requiered: Implement keyword spotting baseline for controlling virtual robot in simulated enviroments (or real robot) via (mobile) app.
  • Example of an ideal result: A student developed an application and extensible library for controlling an e-puck robot both in simulator and physical environment. He demonstrated usability of the the software by using the library API in three tasks. Each of the task was tested by five users. In the first task users were asked to navigate the robot through a maze. In second task two robots were instructed by single user to follow same paths,... In addition to handcrafted baseline a simple logistic regression was used for keyword spotting which obtained better results on small tested with 100 games played by five users.
  • Convenient for any type of project. Complexity is in effectiveness and level of machine learning used.

Dataset and Voice Application for Users with Speech Disorder

  • Required: Selecting dataset and successful application based on literature review, collecting Czech data for evaluation, and implementing simple demonstration application.
    • Note that his topic requires communication skills, and communication and empathy to disabled people is expected;-)
  • Example of an ideal result: Student in his/her thesis reviewed literature, identified convenient sources of annotated data and trained limited vocabulary speech recognition using standard tools which were provided to him/her. Using Czech dataset which was collected during the project he analysed the results and slightly improved the results by adding portion of the collected dataset. The dataset and an example application were published to be used for further research.
  • Convenient as a master thesis or bachalor thesis if data collection is omitted/reduced.

Evaluation and Visualisation Tool for Siri/Google Now/Alex

  • Required: The application will offer navigation in dialogues, trascribing audio from the dialogue and replaying dialogues.
  • Required platform: HTML5 frontend, Python/C# backend
  • Example of an ideal result: A student implemented web-based annotation and evaluation tool which allows to test and compare dialogue systems by replaying part of dialogues. The annotation tools is now used to evaluate and rapidly correct errors in SDS Alex. It was also used to compare dialogue systems services Google Now and Siri.
  • Convenient as master thesis or as a Bachalor thesis/ročníkový projekt if evaluation is greatly reduced.

Taken topics

Debugging of Neural Networks through better visualisation of learning [Specification in progress by Petr Bělohlávek]

  • Required: Test various models and architectures
  • Required: Come up with tasks for evaluation
  • ToBeSpecified ....

Parametric Text-to-Speech (TTS)

  • IMPORTANT: The results will be evaluated in industry.  There is expected financial reward if the implementation is close to state-of-the-art/usable.
  • Details on demand/Coming soon

Summer 2016

Selected Bibliography

All of my publications and talks can be found in biblio

I publish random technical tips on my blog http://opla.cz.