4th Year Project Ideas

Introduction

The following is a selection of a variety of final year projects that I would like to either fully or jointly supervise. They are not an exclusive selection in that if you have an idea that you think I might be suitable to supervise, please feel free to mail me so that we can arrange a discussion.

Speech Transformation for Automatic Pronunciation Tutoring

1- or 2-person project

Summary:

Over the past 2 academic years 2 final year projects have taken vowel sounds pronounced by a sample of native speakers and used these as reference data in an automatic pronunciation tutoring system. The user, i.e. the pronunciation learner, provides a test vowel which is subsequently processed in 2 ways:

1. The user's height and gender are used as parameters to compensate for varying vocal tract lengths which is a factor in spectral variation of vowels. On production of the test utterance, the system provides the user with visual feedback on how "native" the test vowel sound is.
2. The system decomposes the test utterance and synthesises an estimate of how the user should sound in the target language.

The proposed project will expand this system in one or more ways, e.g.:

• Incorporate prosodic (intonational and/or durational) tutoring;
• Incorporate analysis/synthesis of other sound types, which would allow processing of words, phrases, and possibly complete sentences.
• Perform pronunciation variation modelling within a word recognition component that would detect mispronunciations

Prerequisites:

Speech Processing 2

Coverage:

Speech Analysis; Speech Waveform Generation; Maths (particularly linear algebra)

Adaptive Speech Synthesis

1- or 2-person project

Summary:

The idea is to apply speaker adaptation techniques (as used in speech recognition) to speech synthesis. The more data we aquire from a new speaker, the more we can transform an existing synthesiser's voice to sound like the new speaker.

Prerequisites:

Speech Processing 1 & 2; no fear of maths

Coverage:

ASR; Speaker Adaptation; Speech Transformation; Speech Synthesis

-person project

Co-supervisor:

Summary:

Prerequisites:

Coverage:

Articulatory-Gesture Augmented Speech Recognition

2-person project

Co-supervisor:

Alistair Sutherland

Summary:

It has been shown that visual cues such as vocal articulatory gestures can improve speech recognition accuracy. This project would investigate the incorporation of these visual cues to augment a basic speech recogniser

Prerequisites:

Speech Processing 1; Digital Signal Processing

Coverage:

Digital Video Processing, Speech Recognition

"True-Speaker" Dubbing

2-person project

Co-supervisor:

Alistair Sutherland

Summary:

The idea is to use audio-visual synthesis to create a visual image of a speaker speaking a language they have never spoken using speech and visual data from the speaker speaking their native tongue. Concatenative synthesis would be the driving technology for both the speech and visual synthesis. A final year project during 2001/2 has seen some successful preliminary implentation of the speech synthesis side of this project.

Prerequisites:

Speech Processing; Digital Signal Processing

Coverage:

Audio-Visual Synthesis

Text-to-Sign-Language-Gesture Synthesis

2-person project

Co-supervisors:

Alistair Sutherland and Andy Way

Summary:

Deaf people whose primary mode of communication is sign language, would often prefer to "read" signs rather than written words. The aim of this project is to automatically translate/condense text to a series of symbols from which a series of sign-language gestures would be generated as video. This project has been done in 2003/4 but the gesture synthesis was done using animation. This project would instead involved concatenating gestures, digestures, or sub-gesture units taken from recordings of real people.

Prerequisites:

Speech Processing; Digital Signal Processing; Computational Linguistics; Machine Translation

Coverage:

Text Analysis; Sign Language; Machine Translation; Visual Synthesis

Automatic Phonetic Transcription of Irish and Pronunciation Lexicon Building

2-person project

Summary:

The idea is to modify an existing Irish pronunciation lexicon to describe pronunciations of a dialect speaker. Using bootstrapping by manually transcribing a small amount of data, the system will self adjust to the speaker by iterating through a cycle of phoneme recognition, lexicon modification, letter-to-sound (LTS) rule building, and subsequent weighting of the phoneme recognition phonotactic grammar.

Prerequisites:

Speech Processing 1;

Coverage:

Speech Recognition; Phonetic Transcription; Pronunciation Variation; LTS Rules

Irish Speech Recognition

1- or 2-person project

Co-supervisor:

Someone with a decent knowledge of Irish

Summary:

To build a crude speaker dependent Irish speech recogniser.

Prerequisites:

Some knowledge of Irish; Speech Processing 1

Coverage:

Speech Recognition; Phonetics of Irish

Irish Diphone Synthesis for MBROLA

1- or 2-person project

Summary:

This project would involve the creation of an Irish diphone database and incorporating it into the MBROLA system. MBROLA is not a text-to-speech (TTS) synthesizer, and requires phonetic transcriptions and prosodic values, so the project would not be focusing on textual preprocessing of Irish.

Prerequisites:

Coverage:

Part-of-Speech Tagging for Irish

-person project

Co-supervisor:

Summary:

Prerequisites:

Coverage:

Intonational Modelling of Irish

1- or 2-person project

Co-supervisor:

Summary:

Prerequisites:

Coverage: