Arrangementer‎ > ‎

Præsentation af årets speciale 2017

indsendt 2. okt. 2018 04.01 af Hans-Henrik Kaaber   [ opdateret 2. okt. 2018 08.44 ]

Tid: 23. oktober 2018 kl.17:00 til ca. 19:30.
Sted: Aud. D på Niels Bohr Institutet, Blegdamsvej 17, 2100 København Ø
Vejledende tilmelding på

 Velkommen til sæsonens første medlemsmøde hvor vi præsenterer årets bedste specialer fra 2017.

 Nr. 1:
Anders P.K. Dalskov: “End-to-End Encrypted Cloud Storage: A security analysis of SpiderOak ONE”

We examine in the following thesis the proprietary cloud storage application SpiderOak ONE developed and maintained by the Cloud Storage Provider SpiderOak. In a nutshell, the Cloud Storage Provider claims that, due to the user’s data being encrypted before it leaves the user’s computer, only the user (or someone knowing the user’s password) can access the data. In particular, the Cloud Storage Provider cannot read any of the user’s files. We set out to examine this claim. As the application in question does not provide any kind of source code, and little in way of documentation, we first describe how we reverse engineered the application. We then provide a formal description of the authentication protocols used by the application, how it handles cryptographic keys, file encryption and password changes. Finally, we demonstrate several concrete attacks, which a malicious storage provider can carry out that weakens — or entirely breaks — the confidentiality of the user’s password and thus the confidentiality of the user’s stored data.

We disclosed our findings in a responsible manner and SpiderOak updated their product.


Nr. 2:
Kristian Jagd and Mathias Thulstrup Johansen: ”Speech Recognition Systems with Kaldi: Speech-enabling Applications with an Improved Method for Generating Recognizers”

The aim of this thesis is to simplify the process of building and using speech recognizers with the Kaldi toolkit. This is done by creating a tool, which makes it easier and faster for developers to use the open source, speech recognition toolkit Kaldi.

The thesis presents general background information about speech recognition and the speech recognition toolkit Kaldi. An analysis shows why Kaldi is difficult to use for developers who are new to speech recognition. Based on this information, we derive the requirements specification for our tool. We then describe and discuss how the tool is designed and implemented.

Lastly, the developed tool has undergone a user evaluation. The evaluation indicates that the tool does indeed allow the target users to setup and test a speech recognizer based on a Deep Neural Network (DNN) much faster than when using Kaldi without the tool.


Efter fremlæggelserne er der som sædvanlig lidt at drikke og mulighed for at stille yderligere spørgsmål til foredragsholderne eller til almindelig faglig/social snak.