Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.491672
Title: Voice Morphing
Author: Orphanidou, Christina
ISNI:       0000 0001 3459 831X
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2007
Availability of Full Text:
Full text unavailable from EThOS. Please contact the current institution’s library for further details.
Abstract:
Voice Morphing is the process of modifying a source speaker's speech so that it is perceived as if uttered by a target speaker. The numerous applications of this technology have made it a very popular research topic with most approaches focusing on transforming the spectral envelope of the source speaker to match that of the target speaker. This is usually achieved by estimating a transformation function from parallel training data from the two speakers. However, the current applications suffer from loss of spectral detail during synthesis of the morphed speech, often attributed to the low dimensional representations of the acoustic models used for extracting the acoustic features. We propose a new Multiresolution Voice Morphing Algorithm (MVMA) which uses the Discrete Wavelet Transform (DWT) for modeling the shortterm and long-term acoustic properties of speech and estimates a transformation function at each sub-band by means of a Radial Basis Function (RBF) neural network. The proposed system is evaluated objectively and subjectively using parallel and non-parallel training data. The tests show that the proposed approach effectively transforms speaker identity even though perceivable artifacts still exist in the transformed speech.
Supervisor: Not available Sponsor: Not available
Qualification Name: University of Oxford, 2007 Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.491672  DOI: Not available
Share: