Why using voice recognition to identify refugees is controversial

Germany plans to use voice recognition software to verify the many asylum applications it receives -- but the technology is far from perfect, experts warn.

According to the Federal Office for Migration and Refugees (BAMF), around 60 percent of people who sought asylum in the country in 2016 didn't have identification papers with them.

Now, German authorities are planning to use new voice recognition software to verify the asylum seekers' country of origin, according to a report on Die Welt.

The test, which will begin in two weeks and roll out widely in 2018, aims at analysing and identifying the dialects of people seeking asylum using recorded speech samples.

The software is based on the same voice authentication technology used by banks and insurance companies and would help officers review the applications' sources of origin.

Germany has used speech analysis since 1998 to determine country of origin.

Linguistic experts listen to recorded clips of conversation and can discern dialectic variations such as different names for food.

But the new technology has caused some concerns among experts in Germany.

Linguistic expert Monika Schmid told Deutsche Welle that identifying the region of origin could be extremely difficult.

"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," she said.

"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."

Another expert, computer scientist Dirk Hovy at the University of Copenhagen, told Die Welt that the new system would need the creation of a very accurate and broad database, which is a difficult task.

“Creating a perfect dataset is virtually impossible because language is constantly changing,” he said.