RINCE Speech Laboratory receive 2004 SFI Basic Research Grant

Drs Darragh O'Brien and Ronan Scaife

13 July 2004

RINCE researchers Drs Darragh O'Brien and Ronan Scaife of the RINCE Speech Laboratory have been awarded a 2004 SFI Basic Research Grant for an investigation into "New Distance Measures for Concatenative Speech Synthesis"

Current speech synthesisers assemble utterances by concatenating (splicing) short recordings of human speech. The segments of speech which are selected for joining,vary from sub-word units (e.g. syllables) right up to entire phrases.

Anyone who has used a phone-based computer banking system or a talking clock will have heard the unnatural effects (e.g. intonation) produced when whole words are assembled to form sentences. Similarly, when sub-word units are assembled to form words, audible and annoying artifacts occur at the boundaries between the units.

To produce reasonably natural-sounding and intelligible speech, the synthesiser software searches a large database of recordings to find speech units that match well at their boundaries. The key element in this search is a "distance measure",a mathematical formula which tries to calculate how different two speech units would sound to a human listener at the splicing point, and hence how suitable they would be for concatenation.

To date, no one has come up with a distance measure that accurately and consistently models human auditory perception of speech in this context. The RINCE team of O'Brien and Scaife will work on the search for a better distance metric. Not only will this project contribute to better automatic high quality speech synthesis, but it will also advance our understanding of human hearing.