Replay attack detection

Abstract

A method for detecting a replay attack comprises: receiving an audio signal representing speech; identifying speech content present in at least a portion of the audio signal; obtaining information about a frequency spectrum of each portion of the audio signal in which speech content is identified; and for each portion of the audio signal in which speech content is identified: retrieving information about an expected frequency spectrum of the audio signal; comparing the frequency spectrum of the portion of the audio signal in which speech content is identified with a corresponding expected frequency spectrum; and determining that the audio signal may be caused by a replay attack if a measure of the difference between the frequency spectrum of the portion of the audio signal in which speech content is identified and the corresponding expected frequency spectrum exceeds a threshold level.

Description

Technical Field

[0001] Embodiments described herein relate to methods and apparatus for detecting replay attacks on a voice biometrics system. Background Art

[0002] Voice biometric systems are widely used. In such systems, users train the system by providing samples of their speech during an enrollment phase. In subsequent use, the system is able to distinguish between enrolled users and unenrolled speakers. Voice biometric systems can in principle be used to control access to a wide range of services and systems.

[0003] One way a malicious party attempts to defeat a voice biometric system is to obtain a recording of an enrolled user's speech and play it back in an attempt to impersonate the enrolled user and gain access to services that are intended only for enrolled users.

[0004] This is called a replay attack or a spoofing attack. Summary of the invention

[0005] According to one aspect of the present invention, a method for detecting a replay attack is provided, the method comprising:

[0006] receiving an audio signal representing speech;

[0007] identifying speech content present in at least a portion of the audio signal;

[0008] obtaining information about the frequency spectrum of each portion of the audio signal in which speech content is identified;

[0009] For each portion of the audio signal in which speech content is identified, retrieving information about an expected frequency spectrum of the audio signal;

[0010] comparing the frequency spectrum of the portion of the audio signal in which speech content is identified to a corresponding expected frequency spectrum; and

[0011] If the measure of the difference between the frequency spectrum of the portion of the audio signal in which speech content is identified and the corresponding expected frequency spectrum exceeds a threshold level, it is determined that the audio signal is likely caused by a replay attack.

[0012] The method may further comprise:

[0013] removing the effects of sound channels and / or noise from a received audio signal; and

[0014] When information on the frequency spectrum of each portion in which speech content is recognized in the audio signal is obtained, the audio signal is used after removing the influence of a vocal channel and / or noise.

[0015] Identifying speech content present in at least a portion of the audio signal may include identifying at least one test acoustic category. The at least one test acoustic category may include one or more specific phonemes. The at least one test acoustic category may include fricatives, particularly sibilants; and / or may include plosives.

[0016] Identifying at least one test acoustic category may include identifying an occurrence of the test acoustic category in known speech content. The known speech content may include a pass phrase.

[0017] Comparing the frequency spectrum of the portion of the audio signal in which speech content is identified to the corresponding expected frequency spectrum may include:

[0018] The frequency spectrum of the portion of the audio signal where speech content is identified is compared to a corresponding expected frequency spectrum in a frequency band ranging from 5 kHz to 20 kHz.

[0019] Comparing the frequency spectrum of the portion of the audio signal in which speech content is identified to the corresponding expected frequency spectrum may include:

[0020] The frequency spectrum of the portion of the audio signal where speech content is identified is compared to a corresponding expected frequency spectrum in a frequency band ranging from 20 Hz to 200 Hz.

[0021] Comparing the frequency spectrum of the portion of the audio signal in which speech content is identified to the corresponding expected frequency spectrum may include:

[0022] The frequency spectrum of the portion of the audio signal where speech content is identified is compared to the corresponding expected frequency spectrum in the ultrasonic frequency band.

[0023] Comparing the identified portion of the audio signal with the corresponding retrieved information about the corresponding test acoustic class may include:

[0024] A power level in at least one frequency band of the identified portion of the audio signal is compared to a power level in at least one corresponding frequency band of an expected spectrum of the audio signal.

[0025] The measure of the difference between the identified portion of the audio signal and the corresponding retrieved information about the corresponding test acoustic class may include a power difference greater than 1 dB.

[0026] The method may further comprise:

[0027] performing a speaker identification process on the received audio signal; and

[0028] For each test acoustic class, information about the expected frequency spectrum of the audio signal of the speaker identified by the speaker identification process is retrieved.

[0029] The step of retrieving information about an expected spectrum of an audio signal may include: identifying a location in the audio signal where at least one specific acoustic class is expected to appear; and the step of comparing the spectrum of a portion of the audio signal in which speech content is identified with a corresponding expected spectrum may include: comparing the spectrum at the identified location with the expected spectrum at the identified location.

[0030] Certain acoustic classes may include at least some fricative sounds.

[0031] The method may include determining whether the frequency spectrum at the location where fricatives are expected to occur is characteristic of fricatives.

[0032] The step of identifying locations in the audio signal where at least one specific acoustic class is expected to occur may include identifying locations in a predetermined utterance where the at least one specific acoustic class occurs.

[0033] The predetermined utterance may include a pass phrase.

[0034] According to another aspect of the present invention, a system for detecting a replay attack is provided, the system comprising:

[0035] an input for receiving an audio signal representing speech; and

[0036] A processor, wherein the processor is configured to:

[0037] identifying speech content present in at least a portion of the audio signal;

[0038] obtaining information about the frequency spectrum of each portion of the audio signal in which speech content is identified;

[0039] For each portion of the audio signal in which speech content is identified, retrieving information about an expected frequency spectrum of the audio signal;

[0040] comparing the frequency spectrum of the portion of the audio signal in which speech content is identified to a corresponding expected frequency spectrum; and

[0041] If the measure of the difference between the frequency spectrum of the portion of the audio signal in which speech content is identified and the corresponding expected frequency spectrum exceeds a threshold level, it is determined that the audio signal is likely caused by a replay attack.

[0042] There is also provided a device comprising such a system, wherein the device comprises one of the following: a smartphone, a tablet or laptop computer, a game console, a home control system, a home entertainment system, an in-car entertainment system, or a home appliance.

[0043] According to another aspect of the invention, there is provided a computer program product comprising a tangible computer readable medium storing code for causing a suitable programmed processor to perform the method of the first aspect.

[0044] According to a second aspect, a method for detecting a replay attack is provided, the method comprising:

[0045] receiving an audio signal representing speech;

[0046] identifying at least one portion of speech in the audio signal that is comprised of fricative sounds;

[0047] obtaining information about an amount of energy present at ultrasonic frequencies during the at least one portion of the audio signal; and

[0048] A determination is made based on the information regarding the amount of energy present at the ultrasonic frequency whether the audio signal is likely to be caused by a playback attack.

[0049] The method may further comprise:

[0050] obtaining information about an amount of energy present at ultrasonic frequencies during a second portion of the audio signal in which speech consists of voiced speech; and

[0051] Based on the information about the amount of energy present at the ultrasonic frequency and based on the information about the amount of energy present at the ultrasonic frequency during the second portion of the audio signal, it is determined whether the audio signal is likely caused by a playback attack.

[0052] The method may include:

[0053] Based on a ratio of an amount of energy present at the ultrasonic frequency during the at least one portion of the audio signal and an amount of energy present at the ultrasonic frequency during a second portion of the audio signal, it is determined whether the audio signal is likely caused by a replay attack.

[0054] The method may further comprise:

[0055] obtaining information about an amount of energy present at audible frequencies during the at least one portion of the audio signal; and

[0056] Based on the information about the amount of energy present at ultrasonic frequencies during the at least one portion of the audio signal and based on the information about the amount of energy present at audible frequencies during the at least one portion of the audio signal, it is determined whether the audio signal is likely caused by a playback attack.

[0057] The method may include:

[0058] Based on a ratio of an amount of energy present at ultrasonic frequencies to an amount of energy present at audible frequencies during the at least one portion of the audio signal, it is determined whether the audio signal is likely to be caused by a playback attack.

[0059] The method may include:

[0060] calculating a first ratio of an amount of energy present at ultrasonic frequencies to an amount of energy present at audible frequencies during the at least one portion of the audio signal;

[0061] calculating a second ratio of the amount of energy present at ultrasonic frequencies during a second portion of the audio signal in which speech consists of voiced speech to the amount of energy present at audible frequencies during a second portion of the audio signal; and

[0062] A ratio of the first ratio to the second ratio is calculated.

[0063] According to another aspect, there is provided a system for detecting a replay attack, the system comprising:

[0064] an input for receiving an audio signal representing speech; and

[0065] A processor, wherein the processor is configured to:

[0066] identifying at least one portion of speech in the audio signal that is comprised of fricative sounds;

[0067] obtaining information about an amount of energy present at ultrasonic frequencies during the at least one portion of the audio signal; and

[0068] A determination is made based on the information regarding the amount of energy present at the ultrasonic frequency whether the audio signal is likely to be caused by a playback attack.

[0069] There is also provided a device comprising such a system, wherein the device comprises one of the following: a smartphone, a tablet or laptop computer, a game console, a home control system, a home entertainment system, an in-car entertainment system, or a home appliance.

[0070] According to another aspect, there is provided a computer program product comprising a tangible computer readable medium storing code for causing a suitable programmed processor to perform the method according to the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] For a better understanding of the invention, and to show how it may be put into practice, reference will now be made to the accompanying drawings, in which:

[0072] Figure 1 A smartphone is shown.

[0073] Figure 2 is a schematic diagram showing the form of a smart phone.

[0074] Figure 3 An example of a situation in which a replay attack is being performed is shown.

[0075] Figure 4 is a block diagram showing the functional blocks in a voice biometrics system.

[0076] Figure 5 is a flow chart illustrating a method of detecting a replay attack on a voice biometric system.

[0077] Figure 6 Show Figure 4 part of the system.

[0078] Figures 7 to 9 Examples of comparisons between identified portions of an audio signal and corresponding retrieved information about corresponding test acoustic classes are shown in several situations where the audio signal may be caused by a replay attack. DETAILED DESCRIPTION

[0079] The following description sets forth example embodiments according to the present disclosure. Other example embodiments and implementations will be apparent to those of ordinary skill in the art. In addition, those of ordinary skill in the art will recognize that various equivalent technologies may be used to replace or be combined with the embodiments discussed below, and all such equivalent forms shall be deemed to be encompassed by the present disclosure.

[0080] The methods described herein can be implemented in a wide range of devices and systems, such as mobile phones, audio players, video players, mobile computing platforms, gaming devices, remote controller devices, toys, machines, or home automation controllers or appliances. However, for ease of explanation of one embodiment, an illustrative example will be described in which the implementation occurs in a smartphone.

[0081] Figure 1 A smartphone 10 is shown, which has microphones 12, 12a and 12b for detecting ambient sound. In this example, the microphone 12 is of course used to detect the speech of the user holding the smartphone 10, while the microphones 12a, 12b are provided on the upper part of the side of the smartphone 10 and thus are located in the upper part of the side of the smartphone 10. Figure 1 It is not clearly visible.

[0082] Smartphone 10 is just one example of an electronic device in which the method described herein may be performed. The method may similarly be performed in another portable electronic device such as a tablet or laptop computer, a game console, a home control system, a home entertainment system, a car entertainment system, or a home appliance.

[0083] Figure 2 is a schematic diagram showing the form of the smartphone 10 .

[0084] Specifically, Figure 2 Shown are the various interconnected components of the smartphone 10. It should be appreciated that the smartphone 10 will in fact contain many other components, but the following description is sufficient for an understanding of the present invention.

[0085] therefore, Figure 2 The above mentioned microphone 12 is shown. In some embodiments, the smartphone 10 is provided with a plurality of microphones 12, 12a, 12b, etc.

[0086] Figure 2 Also shown is a memory 14 which may in practice be provided as a single component or as a plurality of components. The memory 14 is provided for storing data and program instructions.

[0087] Figure 2 Also shown is a processor 16, which may also be provided as a single component or as multiple components. For example, one component of the processor 16 may be an application processor of the smartphone 10.

[0088] Figure 2 Also shown is a transceiver 18, which is provided for allowing the smartphone 10 to communicate with an external network. For example, the transceiver 18 may include circuitry for establishing an Internet connection via a WiFi local area network or via a cellular network.

[0089] Figure 2 Also shown is audio processing circuitry 20 for performing optional operations on audio signals detected by microphone 12. For example, audio processing circuitry 20 may filter the audio signals or perform other signal processing operations.

[0090] Figure 2 At least one sensor 22 is also shown. In an embodiment of the present invention, the sensor is a magnetic field sensor for detecting a magnetic field. For example, the sensor 22 may be a Hall effect sensor that is capable of providing separate measurements of magnetic field strength in three orthogonal directions. Other examples of sensors that may be used may include a gyroscope sensor, an accelerometer, or a software-based sensor operable to determine the orientation of the phone, wherein such software-based sensors may be used in conjunction with FaceTime, such as provided by Apple, Inc. TMThe system's software programs operate in combination.

[0091] In this embodiment, the smartphone 10 is provided with voice biometric functionality and control functionality. Thus, the smartphone 10 is able to perform various functions in response to verbal commands from a registered user. The biometric functionality is able to distinguish between verbal commands from a registered user and the same command spoken by a different person. Therefore, certain embodiments of the present invention relate to the operation of a smartphone or another portable electronic device with some voice operability, such as a tablet or laptop computer, a game console, a home control system, a home entertainment system, an in-car entertainment system, a household appliance, etc., wherein the voice biometric functionality is implemented in a device intended to implement the verbal commands. Certain other embodiments relate to a system that implements voice biometric functionality on a smartphone or other device, which transmits the command to a separate device if the voice biometric functionality is able to confirm that the speaker is a registered user.

[0092] In some embodiments, while voice biometric functionality is performed on the smartphone 10 or other device located near the user, the spoken command is transmitted using the transceiver 18 to a remote speech recognition system, which determines the meaning of the spoken command. For example, the speech recognition system may be located on one or more remote servers in a cloud computing environment. A signal based on the meaning of the spoken command is then returned to the smartphone 10 or other local device. In other embodiments, the speech recognition system is also provided on the smartphone 10.

[0093] One attempt to fool voice biometric systems is to play a recording of an enrolled user's voice through a so-called replay or spoofing attack.

[0094] Figure 3 shows an example of a situation where a replay attack is being performed. Figure 3 In the example, the smartphone 10 is provided with voice biometric functionality. In this example, the smartphone 10 is at least temporarily possessed by an attacker who has another smartphone 30. The smartphone 30 has been used to record the voice of a registered user of the smartphone 10. The smartphone 30 is brought close to the microphone inlet 12 of the smartphone 10, and the recording of the registered user's voice is played back. If the voice biometric system is unable to detect that the voice of the registered user that it detects is a recording, the attacker will gain access to one or more services that are intended to be accessible only to the registered user.

[0095] As is well known, smartphones such as smartphone 30 are typically provided with relatively low quality speakers due to size constraints. Therefore, a recording of a registered user's voice played back through such a speaker does not perfectly match the user's voice, and this fact can be used to identify a replay attack. For example, a speaker may have certain frequency characteristics, and if these frequency characteristics can be detected in a speech signal received by a voice biometric recognition system, then the speech signal may be considered to be caused by a replay attack.

[0096] For example, at frequencies below a lower threshold frequency, a speaker may suffer from low-frequency roll-off because the bass response is limited by the size of the speaker diaphragm. The lower threshold frequency may be 200 Hz. In another example, at frequencies above an upper threshold frequency, a speaker may suffer from high-frequency roll-off. Although good quality audio recording equipment, amplifiers, speakers, etc. are designed to exhibit very high fidelity in the audible frequency range (e.g., up to about 20 kHz), even high-quality audio systems typically exhibit high attenuation in their frequency response above about 20 kHz. Since the average person cannot perceive sounds in this frequency range, it is generally not worth it for system designers to provide high-fidelity recording and reproduction of such sounds.

[0097] The magnitude of these effects will be determined by the quality of the speaker. For example, in a high quality speaker, the lower and upper threshold frequencies should be such that there is minimal low or high frequency roll-off in the frequency range typically audible to humans. However, size and cost constraints mean that many commercially available speakers, such as those provided in smartphones such as smartphone 30, do suffer from these effects to some extent.

[0098] Therefore, the speaker may not be able to reproduce ultrasonic and / or near-ultrasonic frequencies well. In these cases, if the ultrasonic and / or near-ultrasonic frequency components of the detected speech signal do not reach the power levels expected for "typical" speech, this may indicate that the detected speech signal is caused by a replay attack.

[0099] However, not all speech contains the same levels of ultrasonic and / or near-ultrasonic frequency components. For example, some consonant groups, such as fricatives and plosives, contain relatively high levels of ultrasonic and near-ultrasonic frequencies. If the speech to which the detected speech signal relates contains a higher-than-normal amount of speech in one or more acoustic categories, such as these acoustic categories, which include relatively large components of ultrasonic and / or near-ultrasonic frequencies, then in the result of a playback attack through a speaker that reduces the levels of ultrasonic and / or near-ultrasonic frequency components, the ultrasonic and / or near-ultrasonic frequencies in the detected speech signal may still be present at power levels similar to the power levels that would be expected for these frequencies in "typical" speech. Therefore, in this situation, the system may be unable to recognize that the speech signal may be caused by a playback attack.

[0100] If the acoustic class of a detected speech signal and the expected frequency characteristics of the acoustic class can be determined, it may be possible to use the frequency characteristics of the detected speech signal to more reliably determine whether the detected signal is likely to be caused by a replay attack.

[0101] Figure 4 is a block diagram showing the functional blocks in a voice biometric recognition system, and Figure 5 is a flow chart illustrating a method of detecting a replay attack on a voice biometric system.

[0102] like Figure 4 As shown in , a microphone 12 (e.g., one of the microphones in the smart phone 10) detects sound, and the sound is passed to the initial processing block 40. The microphone 12 is capable of detecting audible sound and sound in the ultrasonic range. As used herein, the term "ultrasound" (and "ultrasound") refers to sound in the upper part of the audible frequency range and above the audible frequency range. Therefore, the term "ultrasound" (and "ultrasound") refers to sound with a frequency above about 15kHz.

[0103] The initial processing block 40 may, for example, include an analog-to-digital converter for converting a signal received from an analog microphone into digital form, and may also include a buffer for storing the signal. Analog-to-digital conversion involves sampling the received signal at a sampling rate. A sufficiently high sampling rate must be selected so that any frequency components of interest are retained in the digital signal. For example, as described in more detail below, some embodiments of the present invention involve looking at the ultrasonic component of the received signal, for example in the region of 20kHz to 30kHz. As is well known from the Nyquist sampling theorem, the sampling rate of a digital signal should be at least twice the highest frequency component of the signal. Therefore, in order to correctly sample a signal containing components at frequencies up to 30kHz, the sampling rate should be at least 60kHz.

[0104] A typical smartphone may be capable of sampling at rates up to 192 kHz (high enough to provide a 96 kHz bandwidth), and may therefore set the sampling rate to any suitable value to accommodate the expected frequency components.

[0105] As is conventional, the signal may be divided into frames, for example having a duration of 10 ms to 100 ms.

[0106] Therefore, in Figure 5 In step 60 of the method, an audio signal representing speech is received.

[0107] The received audio signal representing speech may then be passed to a spectrum extraction block 42. The spectrum extraction block 42 may be configured to obtain a spectrum of the received audio signal. In some examples, the spectrum extraction block 42 may be configured to obtain a power spectrum of the received audio signal, while in some other examples, the spectrum extraction block 42 may be configured to obtain an energy spectrum of the received audio signal.

[0108] In some examples, spectrum extraction block 42 may be configured to perform a Fast Fourier Transform on the received audio signal. The result of the Fast Fourier Transform indicates the power or energy present in the signal at different frequencies.

[0109] In another example, the spectrum extraction block 42 may be configured to apply several bandpass filters to the received audio signal representing speech. Each bandpass filter may only allow signals within a specific frequency band of the received audio signal to pass.

[0110] The received audio signal representing speech and the spectrum obtained by the spectrum extraction block 42 can then be passed to the channel / noise compensation block 44. As will be appreciated, the received audio signal representing speech may be affected by the properties of the channel (which refers to any change in factors that produce a difference between the user's speech and the speech signal generated by the microphone). The received audio signal representing speech may also be affected by noise. The channel / noise compensation block 44 can be configured to remove the effects of the channel and / or noise from the received audio signal representing speech to obtain a received purified audio signal representing speech. An example implementation of the channel / noise compensation block 44 is described in more detail below.

[0111] Specifically, Figure 6 A vocal tract / noise compensation system is shown and is described in more detail in UK Patent Application No. 1719734.4.

[0112] exist Figure 6In the example of FIG. 1 , the signal received on input 70 is passed to a voiced / unvoiced detection block 72 which determines when the received signal represents voiced speech and when it represents unvoiced speech. Thus, the voiced / unvoiced detection block 72 outputs a first component S of the audio signal representing voiced speech. v and the second component S representing unvoiced speech u More specifically, in some embodiments, the first component S of the audio signal representing voiced speech v and the second component S representing unvoiced speech u It is the average spectrum of the voiced and unvoiced components of speech. The average spectrum refers to the spectrum of speech obtained and averaged over multiple frames.

[0113] The signal received on input 70 is also passed to a speaker identification block 74 which performs a speech biometrics process to identify a speaker from a plurality of enrolled speakers. The enrolment process in a speaker identification system typically involves the speaker providing a speech sample, extracting certain features from the speech sample, and using the extracted features to form a model of the speaker's speech. In use, corresponding features are extracted from the speech sample and compared to a previously obtained model to obtain a measure of the likelihood that the speaker is a previously enrolled speaker.

[0114] exist Figure 6 In the system shown in FIG. 7 , one or more speaker models are stored, for example, in a database 76. One or more speaker models are selected based on the output of the speaker identification block 74. In this embodiment, each speaker model contains separate models for voiced speech and unvoiced speech of an enrolled user. More specifically, the model for voiced speech and the model for unvoiced speech of an enrolled user each include amplitude values ​​corresponding to a plurality of frequencies.

[0115] Therefore, the model of voiced speech can be indicated as M v , where M v represents a vector including an amplitude value at each of the selected frequencies, and the model of unvoiced speech can be indicated as M u , where M u Represents a vector containing one amplitude value at each of the selected frequencies.

[0116] As will be appreciated, the received signal containing the user's speech will be affected by vocal tract properties, meaning any changes that produce the difference between the user's speech and the speech signal generated by the microphone, and will also be affected by noise.

[0117] Therefore, assuming that the vocal tract and the noise are constant during the time period during which the received signal is averaged to form the first and second components of the received speech, these first and second components may be expressed as:

[0118] Sv = αMv + n, and

[0119] Su=αMu+n,

[0120] in

[0121] α represents the spectrum of the multiplicative interference component, referred to herein as the channel, and

[0122] n represents the frequency spectrum of the additive interference component, referred to herein as noise.

[0123] Therefore, using the measurements Sv and Su and the models Mv and Mu, the two equations can be solved in comparison block 78 to obtain the two unknowns α and n.

[0124] Therefore, for illustrative purposes,

[0125] and

[0126]

[0127] For the sake of completeness, it should be noted that by making spectral measurements at multiple frequencies, these two equations can be effectively solved at each of the frequencies.

[0128] Alternatively, by measuring at f different frequencies, the equations Sv=αMv+n and Su=αMu+n can each be considered as f different equations to be solved.

[0129] In this case, after solving the equation, it may be useful to apply a low pass filter or a statistical filter (such as a Savitsky-Golay filter) to the result in order to obtain a low pass filtered version of the vocal tract and noise characteristics.

[0130] As an alternative example, the method of least squares may be used to obtain solutions to 2f different equations.

[0131] Thus, information about the channel and / or the noise affecting the audio signal is obtained.

[0132] While one possible implementation of the channel / noise compensation block 44 has been described above, it will be appreciated that the channel / noise compensation block 44 may be configured to perform any suitable method to remove the effects of the channel and / or noise from the received audio signal.

[0133] In addition, although Figure 4The system shown in FIG. 4 includes a channel / noise compensation block 44 (and as described below, the audio signal is then used after the effects of the channel and / or noise are removed), but it will be appreciated that this is optional and the received audio signal can be used directly without removing the effects of the channel and / or noise from the received audio signal. In particular, this can be done in situations where it is known that the channel is approximately ideal or in situations where it is known that the system is in an environment with a low background noise level.

[0134] The effects of the channel and / or noise can then be removed from the input signal, and the output of the channel / noise compensation block 44 can then be passed to a recognition block 46. The recognition block 46 can be configured to identify the speech content present in at least a portion of the received audio signal. Generally speaking, this not only involves detecting the presence of human speech, but also means identifying specific speech content. In some embodiments, the recognition block 46 is configured to identify at least one test acoustic category in the received audio signal, wherein the acoustic category is a phoneme set belonging to the same broad phoneme category, such as fricatives, vowels, etc. In other words, the recognition block 46 can be configured to identify the part of the received audio signal that contains one or more specific test acoustic categories.

[0135] At least one of the tested acoustic categories may be an acoustic category that includes certain spectral properties. In other words, at least one of the tested acoustic categories may be selected to provide a specific power distribution across a specific frequency range. For example, at least one of the acoustic categories may include relatively high levels of ultrasonic and / or near-ultrasonic frequencies.

[0136] In some examples, at least one test acoustic category may include one or more specific phonemes. In some examples, the test acoustic category may include a set of vowels. In other examples, the test acoustic category may include a set of consonants. For example, at least one test acoustic category may include fricatives, and more specifically, sibilants. In another example, at least one test acoustic category may include plosives. It should be noted that the audio signals generated by the human vocal tract, particularly when fricative and sibilant phonemes are clearly pronounced, contain a large amount of energy in the ultrasonic region, above about 20 kHz, and even beyond 30 kHz.

[0137] Identification block 46 may be configured to identify a portion of the audio signal that represents at least one test acoustic category. In some examples, identification block 46 may be configured to identify a portion of the audio signal that represents one of a plurality of preselected test acoustic categories. In some examples, identification block 46 may determine the test acoustic category to which the received audio signal corresponds on a frame-by-frame basis. For example, identification block 46 may determine that a first group of received frames corresponds to a phoneme in the fricative category and that a second group of received frames corresponds to a phoneme in the plosive category.

[0138] The recognition of the acoustic class may be performed, for example, by a trained neural network. Alternatively, the recognition of the phonemes contained in the received phrase may be performed by well-known techniques such as Hidden Markov Models.

[0139] In some examples, recognition block 46 includes or is associated with an automatic speech recognition block that can identify speech content present in at least a portion of the received audio signal. After identifying the speech content, recognition block 46 can then, for example, identify portions of the received audio signal that contain one or more specific test acoustic categories.

[0140] In some examples, the recognition block 46 includes a frame-by-frame estimation of the energy in the near-ultrasonic frequencies, where the frames with the highest energy are associated with the fricative phonemes, effectively creating a time map of the occurrence of the fricative phonemes in the phrase. This time map associated with the received audio signal can be aligned with a similar map previously calculated and stored in the database 48. In text-dependent voice biometrics, the expected time mapping of the phonemes can be calculated during the enrollment phase when the user creates their biometric voiceprint by speaking a predefined passphrase. In text-independent voice biometrics, the mapping can be obtained with the help of a phonetic decoder (or acoustic class identifier) ​​applied to the received audio during an authentication attempt.

[0141] For example, the alignment between the expected and actual time mappings of phonemes may be performed using techniques such as the Viterbi algorithm or dynamic time warping.

[0142] The spectrum of the portion of the signal so aligned can be compared to the spectrum of the corresponding acoustic class. In addition, these alignment techniques can provide a measure of the distance between the time pattern of the fricatives in the received signal and the expected time pattern. A distance exceeding a given threshold (meaning that the ultrasonic component of the received signal does not match the high frequency phonetic content of the expected speech) may indicate a spoofing attack.

[0143] For example, in some embodiments of the method, when applied to a smart phone, smart speaker, etc., the smart phone or smart speaker is activated by a registered user speaking a predetermined pass phrase, and the device is thus able to recognize an audio signal containing the pass phrase. For example, the pass phrase may be "Hi Assistant".

[0144] The vowel sounds in the word "Hi" naturally contain higher frequency components than many vowels. Similarly, sibilant consonants contain relatively large amounts of high frequency components. As a result, the recognition block 46 can determine which parts of the audio signal contain the passphrase, i.e., which frames of the audio signal will contain phonemes with a high proportion of high frequency components. In effect, a map of the time positions of the fricative phonemes and / or sibilant phonemes within the passphrase can be created. Thus, specific speech content in these parts of the audio signal can be identified.

[0145] For example, if the pass phrase occupies 1 second of the audio signal, the frame containing the vowel sound in the word "Hi" may appear in the audio signal, for example, between 50 milliseconds and 150 milliseconds of the audio signal. Similarly, the sibilant consonant in the word "assistant" may appear in the audio signal, for example, between 550 milliseconds and 850 milliseconds of the audio signal.

[0146] Thus, frames occurring during these time intervals may be identified.

[0147] Therefore, in Figure 5 In step 62 of the method, speech content present in at least a portion of the audio signal is identified.

[0148] In such Figure 4 In the example shown in , the output of the channel / noise compensation block 44 is passed to the recognition block 46. However, it will be appreciated that the received audio signal representing speech may be passed directly to the recognition block 46 without any processing previously performed to remove the effects of the channel and / or noise. This may occur in a situation where it is known that the channel is approximately ideal, or this may occur in a situation where it is known that the system is in an environment with very low background noise levels.

[0149] In such Figure 4 In the example shown in , the extracted spectrum is passed to the recognition block 46. However, it will be appreciated that the recognition block 46 may receive the received audio signal representing speech in an unprocessed form or in a form maintained in the time domain (rather than in the frequency domain) and recognize the speech content (e.g., recognize at least one acoustic class) present in at least a portion of the audio signal from the received audio signal in this form. Such recognition may be performed using a suitably trained neural network.

[0150] Having identified the speech content, information is obtained about the frequency spectrum of each portion of the audio signal in which the specific speech content was identified. For example, while the spectrum extraction block 42 may be configured to obtain the spectrum of the entire received audio signal, the second spectrum extraction block 47 may be configured to obtain the spectrum of those portions of the received audio signal in which the specific speech content of interest was identified. For example, as described above, portions of the signal representing specific speech content having a high proportion of high frequency components may be considered to be of interest, and the second spectrum extraction block 47 may be configured to obtain the spectrum of frames of the received audio in which the speech content was identified.

[0151] The recognition block 46 may transmit an indication of the recognized speech content to the database 48. The database 48 may receive the indication of the recognized speech content and supply corresponding stored information about the expected frequency spectrum of the audio signal. For example, where the relevant speech content includes one or more specific acoustic categories (e.g., one or more phonemes), the database 48 may be configured to store, for each acoustic category, an expected power spectrum of the audio signal corresponding to the acoustic category. The database 48 may then be configured to retrieve, for each acoustic category, an expected energy spectrum of the audio signal corresponding to the acoustic category.

[0152] Therefore, in Figure 5 In step 66 of the method of claim 1, for each portion of the audio signal in which speech content is identified, information about an expected frequency spectrum of the corresponding portion of the audio signal is retrieved.

[0153] The database 48 may transmit the retrieved information about the expected frequency spectrum of the audio signal to the comparison block 50 for each portion of the audio signal in which speech content is identified (that is, for example, for each identified test acoustic class).

[0154] The comparison block 50 may also receive from the second spectral extraction block 47 the spectrum of the frames in the received audio signal in which the same speech content is identified (eg the spectrum of the frame or each spectrum containing at least one test acoustic class).

[0155] Comparison block 50 may then compare the received identified portion of the audio signal (representing at least one test acoustic class) with the received corresponding retrieved information regarding the corresponding test acoustic class.

[0156] In some examples, the comparison block 50 may compare the components of the identified portion in the audio signal with the corresponding retrieved information of the corresponding test acoustic category in the frequency band in the range of 5kHz to 20kHz or in the range above 16kHz. As described above, some speakers may not be able to reproduce ultrasonic and / or near-ultrasonic frequencies well. Therefore, a received audio signal including a test acoustic category with relatively high levels of ultrasonic and / or near-ultrasonic frequencies may be poorly reproduced by the speaker at these frequencies (e.g., in the frequency band in the range of 5kHz to 20kHz, or in the frequency band above 16kHz). Therefore, the spectrum representing the test acoustic category in the received audio signal may be significantly different from the expected spectrum representing the test acoustic category in the audio signal, wherein the test acoustic category includes relatively high levels of ultrasonic and / or near-ultrasonic frequencies when the received audio signal is caused by a replay attack. Therefore, in order to be able to detect a replay attack using this frequency band, it is more preferable to make the test acoustic category include relatively high levels of ultrasonic and / or near-ultrasonic frequencies, and the appropriate acoustic category may be, for example, fricatives, such as sibilants and plosives.

[0157] In some examples, the comparison block 50 may compare components of the identified portion of the audio signal with corresponding retrieved information for a corresponding test acoustic category in a frequency band of 20 Hz to 200 Hz. Some speakers may not be able to reproduce lower frequencies (e.g., frequencies within the 20 Hz to 200 Hz band) well. Therefore, a test acoustic level that includes relatively high levels of these frequencies is likely to be poorly reproduced by the speaker at these frequencies. Therefore, the spectrum representing the test acoustic category in the received audio signal may be significantly different from the expected spectrum representing the test acoustic category in the audio signal, where the test acoustic category includes relatively high levels of these lower frequency components when the received audio signal is caused by a replay attack. Therefore, in order to be able to detect a replay attack using this frequency band, it is more desirable to have the test acoustic category include relatively high levels of frequencies in the 20 Hz to 200 Hz band, and suitable acoustic categories may, for example, be certain vowel sounds. For example, in the International Phonetic Alphabet, the test acoustic category is represented by and The high back rounded vowels and near high back rounded vowels represented have high levels of these low-frequency components, respectively.

[0158] In the International Phonetic Alphabet The represented high front unrounded vowels generally have a relatively low first formant frequency F1 and a relatively high second formant frequency F2, and thus the portion of speech containing such vowel sounds has a relatively high level of low frequencies in the 20 Hz to 200 Hz band and a relatively high level of high frequencies in the 5 kHz to 20 kHz band. Therefore, if portions of speech containing such vowel sounds are identified, it is useful to compare the spectra of these portions with the expected spectra in the corresponding frequency bands in both the 20 Hz to 200 Hz range and the 5 kHz to 20 kHz range.

[0159] In some examples, comparison block 50 may compare the frequency spectrum of the identified portion of the audio signal to a retrieved expected frequency spectrum for the corresponding test acoustic class.

[0160] In one example, the comparison block 50 may compare the identified portion of the audio signal with a corresponding retrieved spectrum for the corresponding test acoustic class, wherein the identified portion of the audio signal has been obtained from an FFT performed on the audio signal received at the spectrum block 42. In another example, the comparison block 50 may compare the identified portion of the audio signal with a corresponding retrieved spectrum for the corresponding test acoustic class, wherein the identified portion of the audio signal has been obtained from a band-passed audio signal as generated by the spectrum block 42.

[0161] In some examples, comparison block 50 may compare a power level in at least one frequency band of the identified portion of the audio signal to a power level in at least one corresponding frequency band of the expected spectrum of the audio signal. In this example, the measure of the difference between the identified portion of the audio signal and the corresponding retrieved information about the corresponding test acoustic class may include a difference in power.

[0162] As explained above, when compared to the power level at a specific frequency (or within a specific frequency range) of the corresponding retrieved information regarding the corresponding test acoustic category, a difference in the power level of the identified portion of the audio signal at that frequency (or within that frequency range) may indicate that the audio signal may be caused by a playback attack.

[0163] In some examples, the comparison block may compare power levels in frequency bands in which it may be assumed that the speaker used to replay the attack will have a poor frequency response. The comparison block 50 may compare the power level of the identified portion of the audio signal in a frequency band in the range of 5kHz to 20kHz with the power level of the expected spectrum of the audio signal in the same frequency band in the range of 5kHz to 20kHz. Additionally or alternatively, the comparison block 50 may compare the power level of the identified portion of the audio signal in a frequency band in the range of 20Hz to 200Hz with the power level of the expected spectrum of the audio signal in the same frequency band in the range of 20Hz to 200Hz. Additionally or alternatively, the comparison block 50 may compare the power level of the identified portion of the audio signal in a frequency band in the range of 20kHz to 30kHz with the power level of the expected spectrum of the audio signal in the same frequency band in the range of 20kHz to 30kHz.

[0164] Therefore, in Figure 5 In step 68 of the method, the frequency spectrum of the portion of the audio signal in which speech content is identified is compared with a corresponding expected frequency spectrum.

[0165] The comparison as generated by comparison block 50 may be transmitted to decision block 52. Decision block 52 may determine whether a measure of the difference between the frequency spectrum of the portion of the audio signal in which speech content is identified and the corresponding expected frequency spectrum exceeds a threshold level. If the measure of the difference exceeds the threshold level, decision block 52 may determine that the audio signal may be caused by a replay attack.

[0166] Therefore, in Figure 5 In step 69 of the method, if the measure of the difference between the frequency spectrum of the portion of the audio signal in which speech content is identified and the corresponding expected frequency spectrum exceeds a threshold level, it is determined that the audio signal may be caused by a replay attack.

[0167] In some circumstances, it may be desirable to retrieve information about expected spectra representing a particular test acoustic class in step 66 of the method, where the information is specific to an individual speaker (rather than obtaining spectra representing the test acoustic class spoken by multiple speakers).

[0168] Thus, in some embodiments, the system may perform a speaker identification process on the received audio signal. In some examples, the received audio signal representing speech may be subjected to a speaker identification process before information about the expected frequency spectrum of the audio signal is retrieved from database 48. The speaker identification process may be performed on the received audio signal. Figure 5 The speaker identification method is well understood by those skilled in the art.

[0169] After identifying the speaker who produced the received audio signal representing speech, the obtained speaker identity may also be communicated to the database 48 .

[0170] In some examples, the expected spectrum corresponding to each acoustic class stored in database 48 will represent the acoustic class as spoken by a large group of speakers. However, in some examples, the expected spectrum corresponding to each acoustic class stored in database 48 will represent the acoustic class as spoken by a specific individual. The specific individual may have been identified through a speaker identification process.

[0171] Thus, for each test acoustic class, the database 48 may retrieve information about the expected frequency spectrum of the audio signal of the speaker identified by the speaker identification process.

[0172] For example, there may be multiple registered users in the system, and each of the multiple registered users may have a model corresponding to their speech stored in database 48. For each registered user, the user's corresponding model may include information about the expected frequency spectrum of the user's audio signal for multiple test acoustic categories.

[0173] In an example where there are two registered users in the system, there may be information about two expected spectra for each test acoustic category stored in the database 48, wherein the first expected spectrum corresponds to the first registered user and the second registered user corresponds to the second registered user. The speaker identification process may identify which of the two registered users produced the received audio signal representing speech. The identity of the speaker may then be used by the database 48 to retrieve information about the expected spectrum corresponding to the identified speaker (for each identified test acoustic category in the audio signal). This "speaker-specific" retrieved information may then be passed to the comparison block 50 and used by the comparison block 50 in substantially the same manner as described above. Since the retrieved information about the expected spectrum of the audio signal will be "speaker-specific", the comparison between the identified portion of the audio signal and the corresponding retrieved information about the corresponding test acoustic category may be more accurate (because both pieces of information to be compared are specific to the speaker). Therefore, the determination that the audio signal may be caused by a replay attack may be more accurate as a result.

[0174] Figure 7 , Figure 8 and Fig. 9 Examples of comparisons between identified portions of an audio signal and corresponding retrieved information about corresponding test acoustic classes in several situations where the audio signal may be caused by a replay attack are shown. In the following examples, the identified portion of the audio signal representing at least one test acoustic class is a power spectrum, and the corresponding retrieved information about the corresponding test acoustic class is also a power spectrum.

[0175] exist Figure 7 , Figure 8 and Fig. 9 In , the identified spectrum of the received signal is indicated by a solid line, whereas the expected spectrum that has been retrieved from the database is indicated by a dashed line.

[0176] exist Figure 7 , it can be seen that at a particular first frequency (as indicated by f1), the identified spectrum 82 and the expected spectrum 84 have substantially similar power levels (as indicated by P1). As the frequency increases, the expected spectrum and the identified spectrum will continue to have substantially similar power levels until a particular second frequency (as indicated by f2) is reached. At this point, the expected spectrum 84 remains at the first power level (as indicated by P2), while the power level of the identified spectrum 82 rolls off to a different second power level (as indicated by P3). This roll-off in power level at higher frequencies, as seen in the identified spectrum 82 but not in the expected spectrum for the particular acoustic class, may indicate that the received audio signal representing speech may be caused by a replay attack.

[0177] For example, if the power level decreases at a relatively constant rate (such as 6 dB / octave) for frequencies above the upper cutoff frequency, it may be determined that the received audio signal representing speech is likely caused by a playback attack. The upper cutoff frequency may be in the range of 15 kHz to 25 kHz. Figure 8 , it can be seen that at a particular first frequency (as indicated by f1), the identified spectrum 86 and the expected spectrum 88 have significantly different power levels. The power level of the identified spectrum 86 at f1 (as indicated by P1) is significantly lower than the power level of the expected spectrum 88 at f1 (as indicated by P3). As the frequency increases, the power level of the identified spectrum gradually increases until a second particular frequency (as indicated by f2) is reached. At frequency f2, the power levels of the identified spectrum 86 and the expected spectrum 88 (as indicated by P2) are substantially the same. This "roll-off" of power levels with decreasing frequencies, as seen in the identified spectrum but not in the expected spectrum, may indicate that the received audio signal representing speech may be caused by a replay attack.

[0178] For example, for frequencies below the lower cutoff frequency, if the power level decreases at a relatively constant rate (such as 6 dB / octave), it can be determined that the received audio signal representing speech may be caused by a playback attack. The lower cutoff frequency may be in the range of 20 Hz to 200 Hz.

[0179] exist Fig. 9 , it can be seen that at frequencies below a specific first frequency (as indicated by f1), the power levels of the identified spectrum 90 and the expected spectrum 92 are substantially the same (as indicated by P1). As the frequency increases, the power level of the identified spectrum 90 begins to deviate from the power level of the expected spectrum 92. In this example, the power level of the expected spectrum begins to decrease at a rate much faster than the power level of the identified spectrum as the frequency increases. At a second specific frequency (as indicated by f2), the power level of the identified spectrum 90 is at a much lower power level (as indicated by P3) than the power level of the expected spectrum 92 (as indicated by P2). This difference in absolute power level at a specific frequency between the identified spectrum and the expected spectrum may indicate that the received audio signal representing speech may be caused by a replay attack. In some examples, a difference in absolute power level exceeding a threshold level may indicate a replay attack. For example, the threshold level may be 1 dB. Therefore, by comparing a signal from a known acoustic class with an expected spectrum specific to the same acoustic class, this allows a signal that may be caused by a replay attack to be identified in a more reliable manner.

[0180] Therefore, as mentioned above, one way to detect the limitations of a recording and playback device such as a smartphone and therefore detect when a detected speech signal may be caused by a playback attack is to check the amount of energy present at ultrasonic frequencies. However, the expected amount of energy present at ultrasonic frequencies will depend on the phonemes present in the speech.

[0181] Thus, some methods disclosed herein involve identifying speech content present in at least a portion of an audio signal, and in particular, identifying one or more specific phonemes present in the speech, and then obtaining information about the frequency spectrum of these portions of the audio signal.

[0182] An example of such an approach is to measure the average amount of energy present at ultrasonic frequencies during each frame containing fricative phonemes (or more generally, during each frame containing unvoiced speech) and compare that average amount to the average amount of energy present at ultrasonic frequencies during each frame containing voiced speech.

[0183] therefore,

[0184] in:

[0185] x1 is the measure of speech presence,

[0186] E u (f) is a measure of the amount of energy present at ultrasonic frequencies during frames containing fricative phonemes (or more generally, during frames containing unvoiced speech), and

[0187] E u (v) is a measure of the amount of energy present at ultrasonic frequencies during frames containing voiced speech.

[0188] Thus, a value for x1 may be calculated and compared to an appropriate threshold to determine, with a desired confidence level, whether the speech signal should be considered to be caused by a replay attack.

[0189] Another example of this approach is to measure the average amount of energy present in multiple different ultrasonic frequency bands during each frame containing fricative phonemes (or more generally, during each frame containing unvoiced speech), and compare these average amounts to the amount of energy present in these ultrasonic frequency bands during frames containing voiced speech.

[0190] therefore,

[0191] in:

[0192] x2(b) is a set of values ​​obtained by calculating the following ratios across multiple frequency bands,

[0193] E b (f) is a measure of the amount of energy present in one of several ultrasonic frequency bands during frames containing fricative phonemes (or more generally, during frames containing unvoiced speech), and

[0194] E b (v) is a measure of the amount of energy present in one of several ultrasonic frequency bands during a frame containing voiced speech.

[0195] Therefore, for each of several ultrasonic frequency bands, values, and these values ​​can be combined to form a set of values ​​x2(b), which can then be used as a measure of presence.

[0196] Machine learning methods may be applicable to determine whether the calculated value x2(b) indicates live speech, or whether the speech signal should be considered to be caused by a replay attack.

[0197] Another example of this approach is to measure the average amount of energy present at ultrasonic frequencies during each frame containing fricative phonemes (or more generally, during each frame containing unvoiced speech) and compare that average to the amount of energy present in the audible frequency band during the same frames.

[0198] therefore,

[0199] in:

[0200] x3 is the measure of speech presence,

[0201] E u (f) is a measure of the amount of energy present at ultrasonic frequencies during frames containing fricative phonemes (or more specifically, during frames containing sibilant phonemes, or more generally, during frames containing unvoiced speech), and

[0202] E a (f) is a measure of the amount of energy present in the audible frequency band during the same frame.

[0203] Thus, a value for x3 may be calculated and compared to an appropriate threshold to determine, with a desired confidence level, whether the speech signal should be considered to be caused by a replay attack.

[0204] Another example of this method is to measure the average amount of energy present at ultrasonic frequencies during each frame containing fricative phonemes (or more generally, during each frame containing unvoiced speech) and compare that average amount to the amount of energy present in the audible frequency band during the same frames, and also measure the amount of energy present at ultrasonic frequencies during frames containing voiced speech and compare that amount to the amount of energy present in the audible frequency band during the same frames, and then take the ratio of the two ratios.

[0205] therefore,

[0206] in:

[0207] x4 is the measure of speech presence,

[0208] E u (f) is a measure of the amount of energy present at ultrasonic frequencies during frames containing fricative phonemes (or more generally, during frames containing unvoiced speech),

[0209] E a (f) is a measure of the amount of energy present in the audible frequency band during frames containing fricative phonemes (or more generally, during frames containing unvoiced speech),

[0210] E u (v) is a measure of the amount of energy present at ultrasonic frequencies during frames containing voiced speech, and

[0211] E a (v) is a measure of the amount of energy present in the audible frequency band during frames containing voiced speech.

[0212] Thus, a value for x4 may be calculated and compared to an appropriate threshold to determine, with a desired confidence level, whether the speech signal should be considered to be caused by a replay attack.

[0213] Another example of this approach is to measure the average amount of energy present at ultrasonic frequencies during each frame containing fricative phonemes (or more generally, during each frame containing unvoiced speech), and compare that average amount to the amount of energy present in frames that do not contain speech, which can be considered to be the ultrasonic noise floor.

[0214] therefore,

[0215] in:

[0216] x5 is the measure of speech presence,

[0217] E u (f) is a measure of the amount of energy present at ultrasonic frequencies during frames containing fricative phonemes (or more generally, during frames containing unvoiced speech), and

[0218] E u (n) is a measure of the amount of energy present in the non-speech frame.

[0219] Thus, a value for x5 may be calculated and compared to an appropriate threshold to determine, with a desired confidence level, whether the speech signal should be considered to be caused by a replay attack.

[0220] Those skilled in the art will recognize that some aspects of the above-described apparatus and methods may be embodied as processor control code, such as processor control code on a non-volatile carrier medium (such as a disk, CD-ROM or DVD-ROM), a programmed memory (firmware) such as a read-only memory, or a data carrier (such as an optical or electrical signal carrier). For many applications, embodiments of the present invention will be implemented on a DSP (digital signal processor), an ASIC (application specific integrated circuit) or an FPGA (field programmable gate array). Thus, the code may include conventional program code or microcode, or, for example, code for setting up or controlling an ASIC or FPGA. The code may also include code for dynamically configuring a reconfigurable apparatus (such as a reprogrammable logic gate array). Similarly, the code may include code for a hardware description language such as Verilog TM or VHDL (very high speed integrated circuit hardware description language). As those skilled in the art will appreciate, the code may be distributed between multiple connected components that communicate with each other. Where appropriate, embodiments may also be implemented using code running on a field (re)programmable analog array or similar device to configure analog hardware.

[0221] It should be noted that, as used herein, the term "module" is used to refer to a functional unit or block that may be implemented at least in part by dedicated hardware components such as custom circuits and / or at least in part by one or more software processors or appropriate code running on a suitable general-purpose processor, etc. A module may itself include other modules or functional units. A module may be provided by multiple components or sub-modules that are not necessarily co-located and may be provided on different integrated circuits and / or run on different processors.

[0222] Embodiments may be implemented in a host device, particularly a portable and / or battery-powered host device such as a mobile computing device, for example, a laptop or tablet computer, a game console, a remote control device, a home automation controller or a household appliance, including a household temperature or lighting control system, a toy, a machine such as a robot, an audio player, a video player, or a mobile phone (e.g., a smart phone).

[0223] It should be noted that the embodiments mentioned above illustrate rather than limit the invention, and that those skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. The word "comprising" does not exclude the presence of elements or steps other than those listed in a claim, "a" or "an" does not exclude a plurality, and a single feature or other unit may fulfill the functions of several units set forth in a claim. Any reference signs or labels in a claim should not be construed as limiting its scope.

Claims

1. A method for detecting a replay attack, the method comprising: receiving an audio signal representing speech; identifying speech content present in at least a portion of the audio signal; obtaining information about the frequency spectrum of each portion of the audio signal in which speech content is identified; and For each portion of the audio signal where speech content is identified: retrieving information about an expected frequency spectrum of the audio signal; wherein the searching comprises identifying locations in the expected spectrum where at least one specific acoustic class is expected to occur, the at least one specific acoustic class comprising one or more specific phonemes; Comparing the frequency spectrum of the portion of the audio signal where speech content is identified with a corresponding expected frequency spectrum, wherein the step of comparing the frequency spectrum of the portion of the audio signal where speech content is identified with the corresponding expected frequency spectrum comprises: comparing the frequency spectrum at the identified position with the expected frequency spectrum at the identified position; as well as If the measure of the difference between the frequency spectrum of the portion of the audio signal in which speech content is identified and the corresponding expected frequency spectrum exceeds a threshold level, it is determined that the audio signal is likely caused by a replay attack.

2. The method according to claim 1, comprising: removing the effects of sound channels and / or noise from a received audio signal; as well as When the information on the frequency spectrum of each portion of the audio signal in which speech content is recognized is obtained, the audio signal is used after removing the influence of the vocal channel and / or noise.

3. The method according to claim 1, further comprising: performing a speaker identification process on the received audio signal; as well as For each test acoustic class, information about the expected frequency spectrum of the audio signal for the speaker identified by the speaker identification process is retrieved. The method of claim 1 , wherein the particular acoustic class includes at least some fricative sounds.

5. A method according to claim 4, comprising determining whether the frequency spectrum at the location where fricative sounds are expected to occur is characteristic of fricative sounds.

6. The method of claim 1 , 4 or 5 , wherein the step of identifying locations in the audio signal where at least one specific acoustic class is expected to occur comprises: Locations where at least one specific acoustic class occurs in a predetermined utterance are identified. The method of claim 6 , wherein the predetermined utterance comprises a pass phrase.

8. A system for detecting a replay attack, the system comprising: an input for receiving an audio signal representing speech; as well as A processor, wherein the processor is configured to: identifying speech content present in at least a portion of the audio signal; obtaining information about the frequency spectrum of each portion of the audio signal in which speech content is identified; for each portion of the audio signal in which speech content is identified, retrieving information about an expected spectrum of the audio signal, wherein the retrieving includes identifying locations in the expected spectrum where at least one specific acoustic class is expected to occur, the at least one specific acoustic class including one or more specific phonemes; Comparing the frequency spectrum of the portion of the audio signal where speech content is identified with a corresponding expected frequency spectrum, wherein the step of comparing the frequency spectrum of the portion of the audio signal where speech content is identified with the corresponding expected frequency spectrum comprises: comparing the frequency spectrum at the identified position with the expected frequency spectrum at the identified position; as well as If the measure of the difference between the frequency spectrum of the portion of the audio signal in which speech content is identified and the corresponding expected frequency spectrum exceeds a threshold level, it is determined that the audio signal is likely caused by a replay attack.

9. A device comprising the system of claim 8, wherein the device comprises one of the following: a smart phone, a tablet computer, a game console, a home control system, a home entertainment system, an in-car entertainment system.

10. An apparatus comprising the system of claim 8, wherein the apparatus comprises a household appliance.

11. A computer program product comprising a tangible computer readable medium storing code for causing a suitably programmed processor to perform the method of claim 1.

Images