Effect of Error Correction Strategy on Speech Dictation Throughput

The eight participants in this experiment used two different commercially available speech recognition dictation systems to complete a variety of reading transcription tasks. Participants enrolled fully in both systems. They received training in two correction strategies for both systems: multimodal correction (voice plus mouse plus keyboard) and hands-free correction (voice-only), and used both strategies during the experiment. The key findings were: • Both dictation systems were equally accurate. • Throughput (corrected words per minute) was significantly (63%) faster using multimodal correction. • Speaking rates were the same for both systems and correction strategies, averaging around 105-110 utterances (words and commands) per minute. • Correction speeds for the multimodal correction strategy (13.2 seconds per correction) were significantly faster than (a little more than twice as fast as) those for hands-free correction (29.1 seconds per correction). • At the end of the experiment, participants indicated they significantly preferred the multimodal correction strategy.