Seminar at Speech, Music and Hearing:
Exjob presentations in CS: Sound and Music Computing; Database for Power SupplierVoice analysis; Audio system tests; Household power metering
Dennis Johansson, KTH
Niklas Bäckström. Karl Trumstedt, KTH
Three M.Sc. degree project presentations of general interest. Session chair: Olov Engwall.
AbstractName: Dennis Johansson
Title: Real-time analysis, in SuperCollider, of spectral features of electroglottographic signals
This thesis presents tools and components necessary to further develop an implementation of a method. The method attempts to use the non-invasive electroglottographic signal to locate rapid transitions between voice registers. Implementations for sample entropy and the Discrete Fourier Transform (DFT) implemented in the programming language SuperCollider is presented along with tools necessary to evaluate the method and present the results in real time. Since different algorithms have been used both for clustering and cycle separation, a comparison between algorithms for both these steps have also been done.
Automatic Performance Certification Tool for Digital Multiroom Sound Systems
The company Spotify enables hardware companies to produce sound systems that are capable of playing music via Spotify’s music service. The sound systems are then sold to consumers. In order to guarantee a certain level of quality, the products need to be certified, mainly in regards to performance. This is ideally performed automatically, with the aid of some computer software. In this project, a tool was designed and developed to measure the performance of sound systems, especially those consisting of several loudspeakers, connected via a local area network. The measurements are taken by ordering the sound system to play test tones. The test tones are then recorded, identified and timed by a computer program, by applying a cross-correlation technique, which is otherwise used to perform echo-cancellation among other things. The tool provides improvements and extends the functionality of an existing tool used currently by Spotify.
Evaluation of methods for loading and processing of measurement data in Oracle
Sigma Solutions maintains a system for a Swedish electric power supplier that is responsible for collecting and processing measurement data from different meter points in Swedish households. It receives data from several smaller systems collecting data in the country. This central system that Sigma Solutions developed currently has insufficient performance and can barely handle today’s volume of data. Furthermore there is desire to increase the frequency of measurement, which would heavily increase the volumes of data processed by the system.
The system has an Oracle database where all the data is persisted. With this roughly 35 transactions per seconds is achieved, regarded as far too low The questions in focus for this thesis is what aspects in Oracle and in the applications flows can affect performance and what kinds of techniques that might be applied to increase the performance of the system. Hardware is discussed in this paper but no tests were conducted on the same hardware as the production system and none to see how hardware affects the performance.
A prototype based on the original system was developed and different techniques were tested to see how these affected performance. This thesis also describes tools that can be used to find and adjust performance problems in Oracle as well as features and parameters that can be utilized for different situations when building applications.
The initial version of the prototype system managed only to process one transaction per second. But by testing and evaluating different techniques for reading, updating and saving data in Oracle as well as how data should be processed in the application one version of the system could manage over 5000 transactions per second.
For more information, see Web link
13:15 - 15:00
Thursday November 12, 2015
The seminar is held in Fantum.
| Show complete seminar list