Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.522463
Title: Personalising synthetic voices for individuals with severe speech impairment
Author: Creer, Sarah M.
ISNI:       0000 0004 2693 9198
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2010
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Speech technology can help individuals with speech disorders to interact more easily. Many individuals with severe speech impairment, due to conditions such as Parkinson's disease or motor neurone disease, use voice output communication aids (VOCAs), which have synthesised or pre-recorded voice output. This voice output effectively becomes the voice of the individual and should therefore represent the user accurately. Currently available personalisation of speech synthesis techniques require a large amount of data input, which is difficult to produce for individuals with severe speech impairment. These techniques also do not provide a solution for those individuals whose voices have begun to show the effects of dysarthria. The thesis shows that Hidden Markov Model (HMM)-based speech synthesis is a promising approach for 'voice banking' for individuals before their condition causes deterioration of the speech and once deterioration has begun. Data input requirements for building personalised voices with this technique using human listener judgement evaluation is investigated. It shows that 100 sentences is the minimum required to build a significantly different voice from an average voice model and show some resemblance to the target speaker. This amount depends on the speaker and the average model used. A neural network analysis trained on extracted acoustic features revealed that spectral features had the most influence for predicting human listener judgements of similarity of synthesised speech to a target speaker. Accuracy of prediction significantly improves if other acoustic features are introduced and combined non-linearly. These results were used to inform the reconstruction of personalised synthetic voices for speakers whose voices had begun to show the effects of their conditions. Using HMM-based synthesis, personalised synthetic voices were built using dysarthric speech showing similarity to target speakers without recreating the impairment in the synthesised speech output.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.522463  DOI: Not available
Share: