Detection of replay attack
Detecting a replay attack on a voice biometrics system comprises receiving a speech signal; forming an autocorrelation of at least a part of the speech signal; and identifying that the received speech signal may result from a replay attack based on said autocorrelation. Identifying that the received speech signal may result from a replay attack may be achieved by: comparing the autocorrelation with a reference value; and identifying that the received speech signal may result from a replay attack based on a result of the comparison of the autocorrelation with the reference value, or by: supplying the autocorrelation to a neural network trained to distinguish autocorrelations formed from speech signals resulting from replay attacks from autocorrelations formed from speech signals not resulting from replay attacks.
1. A method of detecting a replay attack on a voice biometrics system, the method comprising:
receiving a speech signal;
forming an autocorrelation of at least a part of the speech signal; and
identifying that the received speech signal may result from a replay attack based on said autocorrelation,
wherein forming an autocorrelation of at least a part of the speech signal comprises:
dividing the at least a part of the speech signal into frames of data;
forming a respective frame autocorrelation of each of the frames of data, wherein each frame autocorrelation extends across a range of lag values, and said range of lag values starts at zero lag; and
averaging said frame autocorrelations to form said autocorrelation;
wherein the received speech signal includes voiced speech segments and unvoiced speech segments; and
wherein the method further comprises:
selecting the voiced speech segments of the speech signal as said part of the speech signal; and
forming an autocorrelation of the selected voiced speech segments of the speech signal; and
wherein identifying that the received speech signal may result from a replay attack comprises comparing at least one feature associated with a negative maximum and/or a positive maximum of the autocorrelation of the selected voiced speech segments of the speech signal with a reference value obtained from at least one other autocorrelation.
2. A method as claimed in claim 1 , wherein identifying that the received speech signal may result from a replay attack comprises:
comparing the autocorrelation with a reference value; and
identifying that the received speech signal may result from a replay attack based on a result of the comparison of the autocorrelation with the reference value.
3. A method as claimed in claim 2 , wherein comparing the autocorrelation with a reference value comprises:
comparing at least one feature of the autocorrelation with a reference value obtained from at least one other autocorrelation.
4. A method as claimed in claim 3 , wherein comparing the autocorrelation with a reference value comprises comparing at least one feature from a first range of lag values of the autocorrelation with a reference value obtained from at least one other autocorrelation.
5. A method as claimed in claim 2 , wherein the reference value is an autocorrelation obtained from a user during enrolment.
6. A method as claimed in claim 2 , wherein the reference value is a property of autocorrelations obtained from multiple users.
7. A method as claimed in claim 1 , wherein identifying that the received speech signal may result from a replay attack comprises:
supplying the autocorrelation to a neural network trained to distinguish autocorrelations formed from speech signals resulting from replay attacks from autocorrelations formed from speech signals not resulting from replay attacks.
8. A method as claimed in claim 1 , wherein the received speech signal includes speech segments and non-speech segments, the method comprising:
selecting the speech segments of the speech signal as said part of the speech signal; and
forming an autocorrelation of the selected speech segments of the speech signal.
9. A system for detecting a replay attack on a voice biometrics system, the system comprising:
an input for receiving a speech signal; and
a processor configured for forming an autocorrelation of at least a part of the speech signal and for identifying that the received speech signal may result from a replay attack based on said autocorrelation, wherein the autocorrelation is formed by:
dividing the at least a part of the speech signal into frames of data;
forming a respective frame autocorrelation of each of the frames of data, wherein each frame autocorrelation extends across a range of lag values, and said range of lag values starts at zero lag; and
averaging said frame autocorrelations to form said autocorrelation;
wherein the received speech signal includes voiced speech segments and unvoiced speech segments;
wherein the processor is further configured to:
select the voiced speech segments of the speech signal as said part of the speech signal; and
form an autocorrelation of the select; and
wherein identifying that the received speech signal may result from a replay attack comprises comparing at least one feature associated with a negative maximum and/or a positive maximum of the autocorrelation of the selected voiced speech segments of the speech signal with a reference value obtained from at least one other autocorrelation.
10. A non-transitory computer readable storage medium having computer-executable instructions stored thereon that, when executed by processor circuitry, cause the processor circuitry to perform a method according to claim 1 .