Thesis

Back to list

Evaluation of linear prediction-algorithms for singing analysis and visualisation of a voice quality measure

Authors Bereuter, P., Kraxberger, F.
Year 2019
Thesis Type Audio Engineering project
Topic Audio Signal Processing
Keywords acoustics, analysis of sound
Abstract Among other methods, linear prediction is widely used in the field of speech signal processing. In this project thesis, the method of linear prediction is applied onto sung vocal signals. The focus lies on the categorisation of sung vocal signals regarding their voice quality and on the recognition of sung vowels. The approach underlying the analysis is the source-filter model, where the source signal is the airstream through the glottis (glottal flow) and the filter is the human vocal tract. Different linear prediction methods are compared with respect to their ability of separating source and filter signals. For the evaluation of the algorithms, synthetic signals with fixed parameter sets for different voice qualities are used. These parameters are taken as the ground truth for evaluating the algorithms with the software Matlab. The analysis method which shows the best results is used for the implementation of an audio plug-in using the JUCE-framework. Therefore, the algorithm has to be adapted for block-wise signal processing, enabling real-time analysis of sung vocal signals using glottal and vocal tract parameters.
Supervisors Sontacchi, A.