Infrastructure and methods for forensic audio transcription experiments

Audio recordings provide powerful forensic evidence in criminal trials. The problem is they are often of poor quality, to the extent the jury needs a transcript to understand the content.

The Research Hub for Language in Forensic Evidence (The Hub) aims to develop evidence-based scientific methods for creating reliable transcripts of indistinct forensic audio. This requires collecting and analysing large amounts of empirical data regarding how listeners interpret indistinct audio under various conditions.

The current project aims to develop tailored infrastructure and methodological protocols to enable ongoing experimental research on this topic.

Over the past ten years, many experiments have been conducted, and methods have evolved. We want to consolidate these methods to streamline and expand our ongoing research. To do this, this collaboration will focus on two major requirements:

1. Infrastructure to deploy audio to participants, allow participants to input their transcripts and enable researchers to add metadata to responses. Existing platforms are too restrictive for our purposes. Dedicated infrastructure will enable us to collect and analyse data more efficiently, with a better experience for participants.
2. Streamlined methodology. Our experimental approach has evolved over the last decade. While this work has produced valuable results, the methods now need to be refined and rationalised into efficient protocols.

Recently, the Hub has been working with the Language Testing Research Centre to develop our capabilities for scoring transcripts. This collaboration would further benefit from the increased interdisciplinarity offered by MDAP, for example in platform development, human sciences and data security. There is scope for outputs of the proposed project to be extended to other branches of linguistic research.