| EPSRC Reference: |
EP/C53042X/1 |
| Title: |
Testing Evaluation of Speech Synthesis (TESSa) |
| Principal Investigator: |
King, Professor S |
| Other Investigators: |
|
| Researcher Co-Investigators: |
|
| Project Partners: |
|
| Department: |
Centre for Speech Technology Research |
| Organisation: |
University of Edinburgh |
| Scheme: |
Standard Research (Pre-FEC) |
| Starts: |
01 January 2005 |
Ends: |
30 September 2008 |
Value (£): |
246,817
|
| EPSRC Research Topic Classifications: |
| Vision & Senses - ICT appl. |
|
|
| EPSRC Industrial Sector Classifications: |
|
| Related Grants: |
|
| Panel History: |
|
|
Summary on Grant Application Form |
|
When deciding whether a speech synthesis system is of better quality than a previous version of the same system (or than another system altogether), researchers generally make use of human auditory evaluation. That is, human subjects are asked to listen to samples of synthetic speech produced by the system, and are asked to rate the quality of the samples on some given parameter. Researchers do not, however, have a very good understanding of how listeners go about rating synthetic speech quality, or of what things (other than the actual quality of the synthetic speech, for instance the type of rating task used) might influence the rating that listeners give. As a result, there is currently no reliable method for evaluating the quality of synthetic speech: researchers tend to choose rating methods at random. Unfortunately, studies have shown that the use of an inappropriate rating method for a given situation can lead to very inconsistent rating behaviour (that is, listeners having difficulty agreeing with either their own previous ratings or with ratings of other listeners). When listeners cannot agree, it is difficult to draw any clear-cut conclusions about the quality of the system, and it is impossible to predict how listeners might react to comparable, but different, changes to the system.The aim of proposed project, therefore, is to take the first steps towards developing a reliable method of testing synthetic speech quality. To achieve this goal, we intend to:1) carry out a wide range of human speech perception (listening) tests to gain a more complete understanding of how human listeners perform the complicated task of evaluating the quality of synthetic speech;2) use the knowledge gained from the above speech perception tests to design new, and more appropriate, rating methods for evaluating the quality of synthetic speech;3) demonstrate that the use of more appropriate rating methods will result in more consistent evaluation (that is, listeners will be able to agree more often with both their own rating and with those of others). More consistent evaluation behaviour can then be used to help design ways of computationally modelling (predicting) how a human will perceive synthetic speech: this will make it easier to evaluate synthetic speech automatically, cutting down on the need for time-consuming, and possibly expensive, perceptual experiments.
|
|
Key Findings |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Potential use in non-academic contexts |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Impacts |
|
| Description |
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk |
| Summary |
|
| Date Materialised |
|
|
|
|
Sectors submitted by the Researcher |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
| Project URL: |
|
| Further Information: |
|
| Organisation Website: |
http://www.ed.ac.uk |