Skip to main page content

Workshop 1 & panel discussion: Audio deepfakes

Presented by: Anil Alexander and Finnian Kelly

It is now possible to create highly-naturalistic synthetic speech samples, commonly referred to as audio deepfakes, using publicly-available tools that require no prior technical knowledge. A recent surge in media reports of fraud and disinformation involving audio deepfakes is testament to the capabilities, and potential harms, of such tools. The rise of audio deepfakes naturally raises questions about the role of the forensic audio expert in the detection of such material, as well as the potential impact on forensic speaker comparison.

The workshop will be in two parts: in the first, we will begin with an overview of the current audio deepfake landscape. We will together explore the creation of deepfakes using commercial and publicly-available services and demonstrate some possibilities for the detection of synthetic speech, including approaches based on human perception. We will then challenge you to ‘spot the deepfake' by listening/looking/analysing in any way you choose.

We hope that this first part of the workshop will inspire a lively panel discussion in the second part, in which we will consider the impact of audio deepfakes in the forensic domain, with input from invited panelists as well as an open Q&A.


Workshop 2: The construction of a speaker profile

Presented by: Gea de Jong-Lendle 

This workshop is based on a cold case presented in Aktenzeichen XY, the German version of Crimewatch. It concerned the murder of a 52-year-old man, Mr.A.L., owner of a construction firm. A few weeks before his death he received a phone call from a person sounding like an elderly lady. She needed building work carried out and requested a meeting to discuss the details. Her suggested time and place were unusual: late evening next to a pub on a quiet land road. However, as the project sounded financially interesting, AL agreed to meet her. Unfortunately, as he turned up on the agreed night, he was shot in his car and died a few days later. The investigation team decided to ask the public for help: in the broadcast they played the two voicemails that had been left by the lady on AL’s answerphone.

In this workshop, we will analyse the two voicemails from the case described above, in order to construct a speaker profile. The focus will be on the standard areas: the speaker’s voice, language and speaking manner. In addition, we try to improve the existing transcript. Although most of the speech is in German, it is assumed that there are other languages involved as well.

During the workshop we will particularly focus on these questions:

  • Sex of the caller
  • Age of the caller
  • Dialect of the caller (for those whose who are familiar with German)
  • Is the transcript correct?
  • Voicemail 1 ends abruptly with three syllables. Any ideas what is being said phonetically, which language is spoken and what they could have meant? 
  • Any other features you found linguistically interesting?

As this is a cold case … I cannot offer the “Mystery Solved”-part at the end or award prizes for the best guesses. However, from a linguistic point of view this case is unusual at a number of different levels and has been the motivation for a number of student projects. You may find it interesting to hear about our analysis so far and to contribute to an unsolved case.

Please bring along a playing device (e.g. mobile phone or laptop) and the associated earbuds / headphones. Audio files will be sent in advance to participants who register here.