Speaker authentication
The speaker recognition makes use of the fact that the voice of a people as unique and distinctive as his fingerprint . This means that access to protected areas can be made much more secure and convenient than, for example, by entering a PIN or password .
In general, a distinction is made between caller verification and identification. The verification checks whether a caller is who he claims to be. To do this, the caller must first identify himself, for example by saying his name or customer number. In a second step, the current voice profile is compared with a stored voice sample (1: 1 comparison).
During identification , a voice sample is recognized from a large number of voice profiles (1: n comparison). The user does not need to identify himself, but must have already given a voice sample beforehand. The disadvantage of this method is that it requires a lot of computing power.
In modern authentication methods, certain algorithms ensure that voice samples are neither rejected if they come from the same speaker, nor are they accepted if they come from a different speaker. In the first case, the systems are designed to be sufficiently tolerant so that they can still recognize a speaker if, for example, the vocal tract has changed due to a cold.
In the second case, for example, randomly generated words that a speaker has to repeat ensure that the system cannot be deceived, for example by pre-recorded voices.
In Switzerland , Swisscom has been using speaker authentication for hotlines since 2016 and PostFinance since 2018 . The knowledge magazine Einstein of the Swiss radio and television had unsuccessfully tried in the edition of February 20, 2020 to hack the voice recognition system of PostFinance .
Individual evidence
- ↑ Marc Bürgi: The sensitive photos from Postfinance and Swisscom. In: handelszeitung.ch . April 2, 2019, accessed April 27, 2019 .
- ↑ Einstein (knowledge magazine): The voice: The new fingerprint. In: srf.ch (video on demand). February 20, 2020, accessed February 21, 2020 .