ASR Phonetic Learning and Mixed Acoustic Model
One of today's vocal applications problems is that customers formulate their requests with great variability. Since it is difficult to reliably predict a priori the user formulations, Loquendo developed a software package for detecting, from field data, user formulations that were not foreseen by the application designers.

The main aims of this tool are the individuation of uncovered frequent formulations and the research of additional pronunciation variants for the foreseen formulations. This is particularly useful in case of non-native speakers and regional accents.

What is more, Loquendo has developed a mixed acoustic ASR model for Castilian Spanish and Catalan. The application developer must specify inside a grammar where words have be catalogued only as Castilian or as Catalan.
The accuracy performance is similar to the performance of the monolingual acoustic models.
This permits in bilingual countries to provide efficient speech services, managing both languages at the same time.