Skip to Content

Department of Linguistics

MU-Talk: Concatenation Modules

Robert Mannell

Click here, for an overview of concatenative synthesis in TTS systems.

In the mid 1990's we began experimenting with two separate approaches to concatenation synthesis. One approach used formant parameter diphones and the other approach used channel vocoder parameter diphones.

To date no work has been carried out on the much more common LPC and waveform concatenation approaches. The reasons for this are complex and are partly related to the technology that we were being funded to develop the TTS system for. Another reason, particularly for channel vocoder parameter concatenation, related to other research that we (ie. Mannell, Clark and occasional other collegues) were carrying out on channel vocoder speech intelligibility. A final reason is that these systems are likely to be much more useful in examining the modelling of voice quality, impaired auditory perception of speech and possibly non-prododic aspects of vocal affect.

Details of these two MU-Talk concatenation modules are dealt with in the following articles:-

  1. Formant-parameter Concatenation
    1. Parallel Formant Synthesis-by-Rule and Diphone-Concatenation utilising data extracted from a Speech Database
    2. Formant Parameter Extraction Procedure

  2. Channel Vocoder-parameter Concatenation
    1. MU-Talk: Channel Vocoder-based Diphone Concatenation

Additional Formant Diphone Article

An additional conference paper (ICSLP98) on formant diphones is also available at a separate location on this site:-