Germany to use voice recognition software to analyze refugees’ dialects
German authorities have revealed plans to use automated speech software to analyze the dialects of refugees to verify their place of origin, according to a report by Deutsche Welle.
The Federal Office for Migration and Refugees (BAMF) are planning to start trials of the speech analysis software within two weeks before implementing a full roll out in 2018.
“The idea is to record a separate speech sample from asylum seekers and to carry out an automatic dialect analysis,” said Julian Detzel, of BAMF’s global strategy digitization and IT program management.
The software, which uses the same voice authentication technology that banks and insurance companies rely on, would enable migration officers to examine the applicants’ sources of origin as one of several “indicators”.
Germany has been using speech analysis since 1998 to assess claims of origin. When authorities are uncertain about a refugee’s place of origin, they send recorded clips of conversation with the applicant to a linguistic expert who is able to spot dialectic variations such as the different names of food.
University of Essex professor Monika Schmid, a linguistics expert who specializes in “language attrition” among migrants, said that “identifying the region of origin for anyone based on their speech is an extremely complex task.”
“We have argued that in order to do so reliably, an analyst must have a solid background in linguistic analysis and be able to take into account a wide range of factors. For example, people will adapt the way they speak to the speech patterns of their interlocutors,” said Schmid. “I don’t see how automated software can distinguish whether a person uses a certain word or pronounces it in a particular way because this is part of their own repertoire or because they were primed to do so by the interviewer or interpreter.”
Schmid and her team recently published a research report where native German speakers listened to various speech samples and determined whether the speaker was a native speaker.
All the samples were of native speakers who had lived abroad for a minimum of five years, but the test group identified several of them as being foreign speakers.
“Crucially, in this context, we also did an acoustic analysis of the pronunciation of each speaker (which, I imagine, must be similar to what automated analysis does). Again, for many speakers the analysis identified non-native pronunciation,” Schmid wrote. “So, both humans and machines can easily be wrong, but humans are probably better at realizing this.”
According to a report by German daily Süddeutsche Zeitung, BAMF had 45 experts for a total of 80 languages. The publication reported that one third of analyses confirmed the claimed country of origin. In 2015 411 such analyses took place.