Kjell Gustafson

Evaluating the pronunciation component of text-to-speech systems for English: a performance comparison of different approaches

The automatic derivation of word pronunciations from input text is a central task for any text-to-speech system. For general English text at least, this is often thought to be a solved problem, with manually-derived linguistic rules assumed capable of handling “novel” words missing from the system dictionary. Data-driven methods, based on machine learning of the regularities implicit in a large pronouncing dictionary, have received considerable attention recently but are generally thought to perform less well. However, these tentative beliefs are at best uncertain without powerful methods for comparing text-to-phoneme subsystems. This paper contributes to the development of such methods by comparing the performance of four representative approaches to automatic phonemization on the same test dictionary. As well as rule-based approaches, three data-driven techniques are evaluated: pronunciation by analogy (PbA), NETspeak and IB1-IG (a modified k-nearest neighbour method). Issues involved in comparative evaluation are detailed and elucidated. The data-driven techniques outperform rules in accuracy of letter-to-phoneme translation by a very significant margin but require aligned text-phoneme training data and are slower. Best translation results are obtained with PbA at approximately 72% words correct on a resonably large pronouncing dictionary, compared with something like 26% words correct for the rules, indicating that automatic pronunciation of text is not a solved problem. c 1999 Academic Press

On the Analysis of Prosody in Interaction

The research reported here is conducted within the ongoing research project “Prosodic Segmentation and Structuring of Dialogue”. The object of study in the project is the prosody of dialogue in a language technology framework. The specific goal of our research is to increase our understanding of how the prosodic aspects of speech are exploited interactively in dialogue—the genuine environment for prosody—and on the basis of this increased knowledge to be able to create a more powerful prosody model. In this paper we give an overview of project design and methods and present some tentative findings.

Modelling of Swedish text and discourse intonation in a speech synthesis framework

In this chapter, we present a model for the analysis and synthesis of intonation in spontaneous conversations in Swedish. The model is an enhanced version of the model developed in (1977) and implemented in our text-to-speech synthesis. In our recent work we have developed the model from the perspective of discourse and multi-sentence texts. This takes us out of the restricted one-sentence/utterance analysis and synthesis into the living world of prosody in communication.

Developing the modelling of Swedish prosody in spontaneous dialogue

The main goal of our current research is the development of the Swedish prosody model. In our analysis of discourse and dialogue intonation, we are exploiting model-based resynthesis. By comparing synthesized default and fine-tuned pitch contours for the dialogues under study, we are able to isolate relevant intonation patterns. This analysis of intonation is related to an independent modelling of topic structure consisting of lexical-semantic analysis and text segmentation. Some results from our model-based acoustic analysis are presented, and its implementation in text-to-speech-synthesis is discussed.