Projects for 2002- 2003

The projects here are a mix of topics, some of which are only suitable for engineering students, but many others are good as well for IT students. I hope to augment the list over the next few days.
 
 1. Concatenative Speech synthesis 

This project aims at developing software for speech synthesis using a existing speech segments. The student will be expected to learn the basics of speech model, use speech data for analysis, development of the model parameters and eventual synthesis from the model parameters. The project is an analysis of how and what parameters to consider at the interface between one diphone and another to make the concatenation as smooth as possible. Requirements are subject knowledge, existing speech corpora and software tools.
This area is of high topical interest because of the move towards direct speech machine - human interface.

2. Automatic Segmentation of Speech using Neural Network

This is an ongoing project aimed at building annotated Maltese speech corpora. The first project used the American TIMIT corpus to build a phoneme based neural net. This was adapted in a second project to test Maltese sentences and adapt the American phoneme set to the Maltese phoneme set. The aim of this project is now to use the existing neural net and appropriate software to overcome the present limitations, to automatically segment existing Maltese speech corpora into its phonemic content, to be able to use this annotated corpus to train a Maltese based speech recognition system.

3. Speech Transformation

This project takes the voice of one person, and through the use of proper analysis and optimisation techniques, transforms this voice into another known voice. This type of transformation is essential today where there are corpora of synthesised speech by a person, which need to be added to, but for which the original person who did the audio recording is no longer available. Requirements are aptitiude for maths and for programming.

4. Acoustic Modelling

The continuation of a project started this year on acoustic modelling. A basic program model has been developed and it is intended to continue the development. Acoustic modelling implies the calculation of parameters based on different models and there use to obtain the response at a receiver based in a room. In this way room acoustics can be modelled prior to building. Particularly in this area there can be subsets of work including: (i) A formal analysis of the acoustic properties of Maltese building materials. This could also require input from the Architecture Department, with whom there can be collaboration on this. (ii) The building of a head auralisation model to simulate the binaural model of hearing as opposed to the omnidirectional model of reception.