Noise in Speech-to-Text Voice: Analysis of Errors and Feasibility of Phonetic Similarity for Their Correction

Noise in Speech-to-Text Voice: Analysis of Errors and Feasibility of Phonetic Similarity for Their Correction Hanna Suominen NICTA / Locked Bag 8001, Canberra ACT 2601, Australia The Australian National University University of Canberra hanna.suominen@nicta.com.au Gabriela Ferraro NICTA / Locked Bag 8001, Canberra ACT 2601, Australia The Australian National University gabriela.ferraro@nicta.com.au Abstract In Australian healthcare, failures in infor- mation flow cause over one-tenth of pre- ventable adverse events and are tangible in clinical handover. Regardless of a good verbal handover, anything from two-thirds to all of this information is lost after 3– 5 shifts if notes are taken by hand or not taken. Speech to text (SST) and informa- tion extraction (IE) have been proposed for taking the notes and filling in a handover form with extrapolated evaluations from related studies promising over 90 per cent correctness for both STT and IE. However, this cascading evokes a fruitful method- ological challenge: the severe implications that errors may have in clinical decision- making call for superiority in STT; the cor- rectness percentage measured in a peace- ful laboratory is decreased to 77 by noise in clinical practise; and the STT errors multiply when cascaded with IE. We pro- vide an analysis of STT errors and dis- cuss the feasibility of phonetic similarity for their correction in this paper. Our data consists of one hundred simulated han- dover records in Australian English with STT recognising 73 per cent of the 7, 277 words (1 h 8 min 5 s) correctly. In text rel- evant to the form, 836 unique error types are present. The most common errors in- clude inserting and, in, are, arm, is, a, the, or am (5 ≤ n ≤ 94), deleting is (n = 17), and substituting and, obs are, 2, he with in, also, to, or and she (7 ≤ n ≤ 11), respec- tively. Eighteen per cent of word substitu- tions sound exactly the same as the correct word and 26 per cent have a similarity per- centage above 75. This encourages using phonetic similarity to improve STT. 1 Introduction Fluent information flow is important in any information-intensive area of decision making, but critical in healthcare. Clinicians are responsi- ble for making decisions with even life-and-death impact on their patients’ lives. The flow is de- fined as links, channels, contact, or communica- tion to a pertinent person or people in the organisa- tion (Glaser et al., 1987). In Australian healthcare, failures in this flow are associated with over one- tenth of preventable adverse events (ACS, 2008; ACS, 2012). Failures in the flow are tangible in clinical handover, that is, when a clinician is trans- ferring professional responsibility and account- ability, for example, at shift change (AMA, 2006). Regardless of verbal handover being accurate and comprehensive, anything from two-thirds to all of this information is lost after three to five shifts if no notes are taken or they are taken by hand (Poth- ier et al., 2005; Matic et al., 2011). There is a proposal to use a …

Content truncated to 500 words