Auditory Mirages in The Transcription of Indistinct Covert Recordings

Transcriptions of covert recordings are a major source of evidence in criminal trials. However, the poor quality of recordings often results in unintelligible speech. There is a widespread belief that poor quality covert recordings can be enhanced with appropriate audio filters, but such procedures may introduce distortions or even remove the audio content we aim to enhance. Additionally, transcription is often subject to contextual priming, a bias in speech decoding due to the effect of context and beliefs about the content of the recording. Here, we propose a new methodology for transcribing covert recordings of poor quality. This method was applied in a recent criminal court case and involves selecting the most relevant audio excerpts, asking several transcribers (who have no prior knowledge of the recording's content) to transcribe the content, and matching the transcriptions with linguistic indices (we used the Jaccard index) to determine the correlation among transcriptions. Transcriptions can then be merged to create a "most likely transcription." In our study, 50 transcribers provided transcriptions under all experimental conditions. They transcribed six audio excerpts: five containing potentially court-relevant content (but of poor audio quality) and one containing irrelevant content but of good quality, which served as a control to assess the transcribers’ ability to decode speech. The overlap between the transcriptions of the various experimental excerpts ranged from a minimum of 34% to a maximum of 77%. The overlap for the control excerpt was 89%. The overlapping material was united to create a "most likely transcription," one of which included relevant information for the court, and above all, enabled the creation of a transcription whose quality was judged to be "Beyond Any Reasonable Doubt."