8. Speech-to-Text (Dictation) Systems
Research on traditional Speech-to-Text Interpreting examines the process of inputting text on a keyboard, the influence of different situational factors on that process, and the impact those factors may have on the end result. This research provides information on effectiveness of work practices, suitable job requirements and working conditions, and possible optimisation measures. Also, research findings in traditional Speech-to-Text Interpreting can be partially transferred to Respeaking, where Speech-to-Text Systems are currently employed (cf. Research Area 3).
In Research Area 8, different dictation systems will be catalogued and their properties compiled and compared. Currently, only speaker-dependent speech-to-text systems provide adequate results. These systems must first be trained to fine-tune the recognition of the respeaker’s voice, resulting in increased accuracy (cf. Research Area 3). Furthermore, in all dictation systems any specialist vocabulary must be entered beforehand and its pronunciation trained by the respeaker.
One research question is whether Speech-to-Text (Dictation) Systems can be improved to the extent that speakers can be directly connected to them without system training procedures. Since live subtitles cannot be flawless even in the best-case scenario, an analysis of how people with hearing impairments who regularly use live subtitles evaluate those errors will be undertaken. Linguistics and translation studies will serve as the theoretical frameworks for this analysis.
In Research Area 8, the project partners will coordinate the collection of existing research output in this area. Speech-to-Text and, possibly, Text-to-Speech Systems will also be tested in project works by students of the Department of Applied Linguistics and the School of Engineering of the ZHAW. The findings will then be used to initiate further work in this area.
The main objectives of Research Area 8 are therefore:
- To catalogue existing Speech-to-Text (Dictation) Systems and compare their properties.
- To analyse how students with hearing impairments assess errors in live subtitles.
- To carry out a series of tests with speaker-dependent and speaker-independent systems for Live-Subtitling and other possible applications.