The recent introduction of vocal input systems in diagnostic radiology will produce many changes in reporting. Statistical analyses of the texts put in (the so-called "mediated vocal input") are an interesting possibility which allows the radiologist to evaluate both readability and efficacy of his reports. The structure of a vocal input system is also suitable to define standardized vocabularies of commonly used radiological terms. This paper describes the structure of an original system, built by assembling low-cost components, with a self-made software. The aim is the reduction of noise in the communication between the radiologist and the physician, by means of new solutions which will make the radiological report one of the key points of the diagnostic procedure.