Developing speech recognition applications for market research

Date of publication: September 1, 1994


This paper describes progress to date on a government supported programme to evaluate the potential applications of Automatic Speech Recognition (ASR) to market and survey research. The paper reviews the growth of telephone research in Europe in recent years and goes on to discuss how technology might be used to reduce costs without reducing quality. ASR is briefly described and its potential for self-completion discussed. The paper progresses to describing the manner in which an original large scale and representative English speech database has been built up for use in sub-word modelling. The body of the paper discusses the new and challenging issues of conducting interviewerless telephone interviews. The ways in which software is required to replace the implicit skills of interviewers are outlined and examples given of how these have been tested in the field by replicating an existing continuous customer quality service. Results from the field test are given and an outline of the second field test which will be reported at the Conference. The paper ends by concluding that ASR could be used in market research by interviewers or respondent panels. The extent to which ASR is used will depend very much on whether software tools can be developed which will enable dialogues to be written quickly by researchers rather than speech engineers. This will take time and considerable investment. The result could be a multi-lingual technique capable of interviewing the world from a single site.

  • PDF
  • This could also be of interest