Speech Science Forum - Georgina Brown

Though they certainly overlap, there is some distance between the fields of speech technology and forensic speech science in terms of their size, typical membership and the pace at which they innovate. These observations are not intended to be criticisms of one field or the other, but they are good reasons for forensic speech scientists to identify further opportunities to work with speech technology to make meaningful progress. The main point at which speech technology and forensic speech science intersect is at automatic speaker recognition, as voice identification is the core task of forensic speech science. Particularly if we explicitly broaden the scope of ‘forensic’ speech science to include ‘investigative’ and ‘intelligence’ contexts, the ways in which speech technology can make a contribution go beyond traditional ideas of automatic speaker recognition research. First, this talk provides a rundown of different speech technologies (automatic accent recognition, automatic speech recognition and automatic speaker recognition) that have been considered for forensic speech science activities. Crucially, this talk will offer a frank reflection on the current performance and status of each technology and the prospect of deployment in forensic and investigative settings. This talk then tackles the topic of ‘responsible AI’ and presents, in more detail, a novel way in which speech technology can be used to enhance system “explainability”. Specifically, it presents how Text-to-Speech (TTS) technology can be used to increase the explainability and transparency of state-of-the-art automatic speaker recognition systems. Achieving this could bring huge benefits to forensic speech science.