Socialproof
Shaojian Zhu · 2012
Though various Automatic Speech Recognition (ASR) based text correction techniques have been proposed, it is still difficult to correct dictation errors using speech based commands. Inspired by the successful use of crowdsourcing to solve computation tasks, we propose SocialProof, a crowdsourcing powered ASR dictation enhancement, to provide a powerful and accurate but fairly cheap ASR dictation system. SocialProof begins with the output produced by ASR engines and enhances this output using the power of crowd intelligence via MTurk service. Our system splits one ASR dictation scenario into several smaller tasks, allowing multiple people to work on different pieces of the task at the same time. Data merging strategies are used to combine multiple responses from MTurk workers to provide improved results. An evaluation of SocialProof strongly supports the effectiveness of this approach.