Approved
Automated Error Detection in Audio Signals
Zoltan Michis () and Martin Kilsgård ()
Start
2007-06-01
Presentation
2008-09-26 09:15
Location:
Finished:
2009-03-01
Master's thesis:
(Contact supervisor)
Abstract
In this master thesis, algorithms for detection and pinpointing of errors within an audio stream were developed. The goal for the algorithms was to be able to detect amplitude spikes, undesired pauses and general distortion. Four different algorithms where developed, one for each of the above anomalies and one for error detection in sinusoids. An LPC based error detection algorithm was designed to detect amplitude spikes, undesired pauses and heave distortion in strongly periodic data, such as sinusoids. An second algorithm was designed to detect undesired pauses in a mixed audio content, i.e, music or speech. It is able to distinguish between desired pauses, which are common in speech content, and undesired pauses. An third algorithm was designed to detect amplitude spikes in a mixed audio content, i.e, music or speech. Finally, a PCA based Audio Classifier was designed to detect a wide variety of distortions, both in frequency and amplitude.
Supervisor: Martin Stridh (EIT)
Examiner: Leif Sörnmo (EIT)