Modular engineering prototyping plan for speech recognition in a visual object oriented environment

A speech processing system may require many different processing steps. It is therefore useful to achieve a modularity that can ensure the re-usability of different processing units in different applications. In this paper an engineering approach to quickly create prototypes for speech recognizers based on HMM theory is proposed. The presented system can be used for many different selected speech applications. Here it has been tuned specifically for words and sentences recognition of a regional variety of Italian. Experimental results have been obtained using the AVIP corpus of Italian spontaneous speech. The prototyping sy stem is based on the "Khoros" software framework; it offers the possibility of easily implementing a processing system planned for high software modularity and re-usability. This approach leads to quick modifications and extensions of the planned speech system.