Speech Recognition Faced with a Language Variety: Andalusian Dento-Alveolar Affricate [el Reconocimiento del Habla Ante la Variedad de Una Lengua: La Africada Dento-Alveolar Andaluza]

The quality of voice recognition within the field of speech technologies depends on the type of system used and the application to which it is destined. One of the problems that it can present is its degree of robustness with dialectal variants of a language. In this study, an analysis is done of the recognition rate of Andalusian aspiration before the voiceless dentoalveolar plosive by using Google Voice Search and untrained Windows 7, on the one hand, and Windows 7 after a number of training sessions with a speaker, on the other hand. Initially, Google Voice Search has greater robustness than Windows 7, which Windows 7 compensates as the number of training sessions increase. In general, it can be observed that recognition robustness is affected by factors such as the number of possible lexical candidates and the context of the entries.