PhD position available - modeling acquisition of speech

We are looking for someone who has (or who is about to complete) a master’s degree in artificial intelligence, speech technology, computer science or equivalent. You will work on a project that investigates advanced techniques for learning the building blocks of speech, with a focus on spectro-temporal features and dynamic Bayesian networks. It is part of the Artificial Intelligence lab of the Vrije Universiteit Brussel and is supervised by prof. Bart de Boer.

The PhD project

Speech consists of a limited number of building blocks that are combined into an essentially unlimited number of words. Children acquire the building blocks and the rules to combine them effortlessly. Although modern speech technology, especially that based on deep learning, has made great progress on this task, it still needs much more training data than children do. This PhD project explores the hypothesis that a fundamental problem of speech technology is the separation of spectral (often modeled by MFCCs) and temporal features (often modeled by HMMs). It will investigate an alternative where spectro-temporal features are used, with a learning technique that is based on dynamic Bayesian networks.

The main idea for a concrete implementation of this is to automatically generate a large number of spectro-temporal feature detectors (sensitive to clicks, tones, harmonic stacks, noise bursts, formant patterns etc.) which become more or less active in the course of a speech signal. The dynamic Bayesian network then has to learn which combinations and sequences of features correspond to which phonemes/syllables/words. The hypothesis is based on the conjecture that this method is much closer to human speech recognition than traditional computer speech recognition and that this will result in more robust computer speech recognition, and in more insight into how humans perceive speech.

Requirements

  • Either one of (and preferably both):
    • Knowledge of speech processing
    • Knowledge of Bayesian techniques, especially dynamic Bayesian networks
  • Good programming skills, preferably experience with C++11
  • An affinity with cognitive and linguistic questions is an advantage

Conditions

You apply for a four-year PhD scholarship, the exact amount of which depends on your experience, but which is sufficient to live comfortably in Brussels. There is also some additional money to visit conferences and to hire research assistants. You will work in an international and interdisciplinary research group. The group is part of the computer science department, but it investigates questions about (the evolution of) cognition and language, using a combination of experiment and computer modeling. You will be asked to do a limited amount of teaching and supervision, but that will take up no more than 20% of your working time. As part of your PhD project, you will have the possibility to spend a few months at a different institute.

The envisaged starting date is October 1st, 2017

How to Apply

We need a letter of motivation, a CV, a copy of your master’s thesis (or in case you have not completed this, another sample of your writing), a list of your grades and the contact information of two people who can provide references. Please send these in pdf format to Bart de Boer: bart@arti.vub.ac.be mentioning “DBN PhD” in the subject. For inquiries, you can send a message to this same address, or phone +32 2 629 3755. The deadline for application is July first, 2017.

The Vrije Universiteit Brussel

The Vrije Universiteit Brussel has been a leading player in Flemish Higher Education for 40 years. The University has 10.000 students and together with its hospital employs more than 5.500 people.

The following form part of the University’s mission:

·            the development, the communication and the application of a high level of academic education and scientific research, free from all preconceived ideas;

·            the translation of these ideals and knowledge into society in the spirit of social engagement;

·            the creation of a society in which everyone is capable of engaging in critical thinking.

Working environment

As an employee of the Vrije Universiteit Brussel you work in a dynamic, diverse and multilingual environment. Both our campuses are set within green oases on the outskirts of the centre of the capital of Flanders, Belgium and Europe. This centre, with all its opportunities, is within reach by public transport in less than half an hour.

Depending on your experience and academic merits you will receive a scholarship or salary on one of the pay scales laid down by the government. Health insurance and free use of public transport for travel to and from work are standard conditions of employment. If you would rather cycle to work, compensation is also available for that. Both campuses have extensive sporting facilities that are at your disposal and a nursery is within walking distance.

More information is available at www.vub.ac.be under the heading ‘future employees’.