Interface design strategies for computer-assisted speech transcription

Abstract

A set of user interface design techniques for computerassisted speech transcription are presented and evaluated with respect to task performance and usability. These techniques include error-correction mechanisms which originated in dictation systems and audio editors as well as new techniques developed by us which exploit specific characteristics of existing speech recognition technologies in order to facilitate transcription in settings that typically yield considerable recognition inaccuracy, such as when the speech to be transcribed was produced by different speakers. In particular, we describe a mechanism for dynamic propagation of user feedback which progressively adapts the system to different speakers and lexical contexts. Results of usability and performance evaluation trials indicate that feedback propagation, menu-based correction coupled with keyboard interaction and text-driven audio playback are positively perceived by users and result in improved transcript accuracy.

Efficiency and usability study of innovative computer-aided transcription strategies for video lecture repositories

“…This conclusion differs from that of Luz et al (2008) mainly because a greater amount of adaptation data has been used in our study to effectively perform the adaptation of acoustic and language models.…”

“…In line with Luz et al (2008), more sophisticated user interfaces alone, like our intelligent interaction strategy, were not proven more efficient in terms of WER decrease per RTF unit than conventional post-editing, nor were they preferred by lecturers over the simple (though more time-costly) interaction model. We find it particularly noteworthy how important it was for lecturers to be able to produce high quality (perfect) end transcriptions, prioritising this over any time-savings afforded by the more intelligent strategies (Munteanu et al (2006); Pan et al (2010); Favre et al (2013)): a half of our lecturers reverted to the conventional post-editing model to complete the review of their video transcriptions.…”

“…First, alternative variants of intelligent interaction strategies which, while allowing lecturers full control over transcription quality, are better able to exploit confidence measures and visual representation (Luz et al (2008(Luz et al ( , 2010). Second, the most suitable interface design for transcription review could be determined on a case-by-case basis, perhaps as a function of WER.…”

“…In this work, we expand the preliminary results reported in (Valor Miró et al (2014)) in order to provide an in-depth analysis of a series of more intelligent active user interaction strategies for the generation of transcriptions that are accurate enough to be useful to students while requiring the minimum effort on the part of the lecturer (Luz et al (2008)). To this end, a three-phase evaluation process was set up to analyse alternative user interaction strategies for reviewing the automatically-generated transcription.…”

Qualitative investigation of the display of speech recognition results for communication with deaf people

“…In [10] confidence measures were used to highlight words with low confidence scores in view of helping error correction in a multimodal environment. Along this line, it is always words with low confidence scores that are differentiated, either in a lighter shade for error correction in voicemail transcripts [11], or highlighted for computer assisted speech transcription [12], or displayed with an underlining dependent on the confidence measure [13]. As the confidence measures are not perfect such approaches do not always accelerate the detection and correction of the errors [13].…”