First adopted by radiology more than 25 years ago, speech recognition (SR) technology has been significantly improved. State-of-the-art SR systems are characterized by extensive vocabularies and advanced mathematical language models which theoretically allow for mean recognition rates far beyond 90%. Commercial arguments for SR implementation are a reduction of report turnaround times (RTT) and cost savings by the elimination of transcriptionist services. In contrast, the (potential) negative impact of front-end SR on radiologists' productivity has been discussed extensively in the literature. This study was conducted in a 1400 bed university hospital and aims to retrospectively evaluate the effect of an SR system on report availability and RTT over a 1.5 year period after integration into a preexisting RIS/PACS installation. Front-end SR with editing of the draft report by the radiologist (night and weekend reports) and back-end SR with correction of the draft report (put out by the recognition servers) by transcriptionists (reports generated during routine working time) were utilized.