ETRO VUB
About ETRO  |  News  |  Events  |  Vacancies  |  Contact  
Home Research Education Industry Publications About ETRO

Master theses

Current and past ideas and concepts for Master Theses.

Hey Alexa, do you remember me? – Audio-based Speaker verification

Subject

In the past few years, voice-controlled smart assistants like Alexa (Amazon Echo), Google Assistant (Google Home) and Siri (Apple HomePod) have become very popular in US and UK households, where they are used to play music, keep up with agendas and appointments, control smart devices and even to place online orders. However, because these systems typically accept spoken commands from anyone they hear, some problems have arisen over time:


To avoid these unwanted triggers, Amazon and Google have recently introduced speaker verification algorithms, which can check whether the speaker is someone authorised to use the device. In this thesis, you'll be designing and implementing a similar speaker verification system, based on deep learning and noise-robust speech analysis algorithms that are also used in forensic speaker analysis.

Kind of work

In this thesis, the student will investigate speech analysis algorithms that can be used for speaker verification. In general, the work will consist of the following steps:


  • Literature study on (forensic) speech analysis, speaker verification, and deep learning.

  • Selection of one or more existing speech databases for speaker verification/recognition, to be used for training and testing of the speaker verification system

  • Implementation of the speech analysis algorithms and the speaker verification framework.

  • Training of the speaker verification system, evaluation and selection of the most relevant analysis algorithms.

  • Evaluation of your speaker verification system

  • Thesis report and presentation


Optional: If sufficient time is available, you can combine your speaker verification system with an existing keyword recognizer (based on (Pocket)Sphinx) to implement a speech-based control interface for e.g. smart lights that only listens to your commands.

We expect that most (if not all) implementation work can be performed in Matlab. Depending on your experience, (partial) C or C++ implementations are also possible

Framework of the Thesis

This thesis is related to the speech analysis research track at the audiovisual signal processing group at ETRO.

Number of Students

1

Expected Student Profile

Engineering or informatics student with knowledge of digital signal processing and Matlab (or C/C++).

Promotor

Prof. Hichem Sahli

+32 (0)2 629 2916

hsahli@etrovub.be

more info

Supervisors

Ir. Henk Brouckxon

+32 (0)2 629 3955

hbrouckx@etrovub.be

more info

Dr. Ir. Lukas Latacz

+32 (0)2 629 3955

llatacz@etrovub.be

more info

Image

Google Home

- Contact person

- IRIS

- AVSP

- LAMI

- Contact person

- Thesis proposals

- ETRO Courses

- Contact person

- Spin-offs

- Know How

- Journals

- Conferences

- Books

- Vacancies

- News

- Events

- Press

Contact

ETRO Department

info@etro.vub.ac.be

Tel: +32 2 629 29 30

©2019 • Vrije Universiteit Brussel • ETRO Dept. • Pleinlaan 2 • 1050 Brussels • Tel: +32 2 629 2930 (secretariat) • Fax: +32 2 629 2883 • WebmasterDisclaimer