Audiovisual deepfake detection
The embodiments execute machine-learning architectures for biometric-based identity recognition (e.g., speaker recognition, facial recognition) and deepfake detection (e.g., speaker deepfake detection, facial deepfake detection). The machine-learning architecture includes layers defining multiple scoring components, including sub-architectures for speaker deepfake detection, speaker recognition, facial deepfake detection, facial recognition, and lip-sync estimation engine. The machine-learning architecture extracts and analyzes various types of low-level features from both audio data and visual data, combines the various scores, and uses the scores to determine the likelihood that the audiovisual data contains deepfake content and the likelihood that a claimed identity of a person in the video matches to the identity of an expected or enrolled person. This enables the machine-learning architecture to perform identity recognition and verification, and deepfake detection, in an integrated fashion, for both audio data and visual data.
1. A computer-implemented method comprising:
obtaining, by a computer, an audiovisual data sample containing audiovisual data;
applying, by the computer, a machine-learning architecture to the audiovisual data to generate a similarity score using a biometric embedding extracted from the audiovisual data, generate a lip-sync score using one or more lip-sync embeddings extracted from the audiovisual data, and generate a deepfake score using a speaker spoofprint embedding and a facial spoofprint embedding extracted from the audiovisual data; and
generating, by the computer, a final output score indicating a likelihood that the audiovisual data is genuine based upon algorithmically combining the similarity score, the lip-sync score, and the deepfake score.
2. The method according to claim 1 , further comprising identifying, by the computer, the audiovisual data sample as a genuine data sample in response to determining that the final output scores satisfies a threshold.
3. The method according to claim 1 , further comprising identifying, by the computer, deepfake content in the audiovisual data in response to determining that the deepfake score satisfies a deepfake detection threshold.
4. The method according to claim 1 , wherein the biometric embedding includes at least one of a voiceprint embedding and a faceprint embedding.
5. The method according to claim 1 , further comprising:
extracting, by a computer, a voiceprint embedding for an audiovisual sample by applying a speaker embedding extraction engine of the machine-learning architecture to an audio signal of the audiovisual data; and
extracting, by the computer, the speaker spoofprint embedding for the audiovisual data by applying an audio spoofprint embedding extraction engine of the machine-learning architecture to the audio signal of the audiovisual data.
6. The method according to claim 1 , further comprising:
extracting, by a computer, a faceprint embedding for the audiovisual data by applying a faceprint embedding extraction engine of the machine-learning architecture to visual media of the audiovisual data; and
extracting, by a computer, the facial spoofprint embedding for the audiovisual data by applying a visual spoofprint embedding extraction engine of the machine-learning architecture to the visual media of the audiovisual data.
7. The method according to claim 1 , further comprising extracting, by the computer a feature for a speaker voiceprint embedding of the biometric embedding, the feature extracted from audio data of the audiovisual data.
8. The method according to claim 1 , further comprising extracting, by the computer, a feature for a faceprint embedding of the one or more biometric embeddings, the feature extracted from image data of the audiovisual data.
9. The method according to claim 1 , further comprising parsing, by the computer, the audiovisual sample into a plurality of segments having a preconfigured length, wherein the computer generates the biometric embedding and at least one spoofprint embedding for each segment.
10. The method according to claim 1 , further comprising generating, by the computer, a lip-sync score by applying a lip sync estimation engine of the machine-learning architecture on the audiovisual data, wherein the computer generates final output score using the lip-sync score.
11. The method of claim 1 , wherein audiovisual data contains audio data, image data, or both audio data and image data.
12. A system comprising:
a computer comprising a processor configured to:
obtain an audiovisual data sample containing audiovisual data;
apply a machine-learning architecture to the audiovisual data to generate a similarity score using a biometric embedding extracted from the audiovisual data, generate a lip-sync score using one or more lip-sync embeddings extracted from the audiovisual data,—and generate a deepfake score using a speaker spoofprint embedding and a facial spoofprint embedding extracted from the audiovisual data; and
generate a final output score indicating a likelihood that the audiovisual data is genuine based upon algorithmically combining the similarity score, the lip-sync score, and the deepfake score.
13. The system according to claim 12 , wherein the computer is further configured to identify the audiovisual data sample as a genuine data sample in response to determining that the final output scores satisfies a threshold.
14. The system according to claim 12 , wherein the computer is further configured to identify deepfake content in the audiovisual data in response to determining that the deepfake score satisfies a deepfake detection threshold.
15. The system according to claim 12 , wherein the biometric embedding includes at least one of a voiceprint embedding and a faceprint embedding.
16. The system according to claim 12 , wherein the computer is further configured to:
extract a voiceprint embedding for an audiovisual sample by applying a speaker embedding extraction engine of the machine-learning architecture to an audio signal of the audiovisual data; and
extract the speaker spoofprint embedding for the audiovisual data by applying an audio spoofprint embedding extraction engine of the machine-learning architecture to the audio signal of the audiovisual data.
17. The system according to claim 12 , wherein the computer is further configured to:
extract a faceprint embedding for the audiovisual data by applying a faceprint embedding extraction engine of the machine-learning architecture to visual media of the audiovisual data; and
extract the facial spoofprint embedding for the audiovisual data by applying a visual spoofprint embedding extraction engine of the machine-learning architecture to the visual media of the audiovisual data.
18. The system according to claim 12 , wherein the computer is further configured to extract a feature for a speaker voiceprint embedding of the biometric embedding, the feature extracted from audio data of the audiovisual data.
19. The system according to claim 12 , wherein the computer is further configured to extract a feature for a faceprint embedding of the biometric embedding, the feature extracted from image data of the audiovisual data.
20. The system according to claim 12 , wherein the computer is further configured to parse the audiovisual sample into a plurality of segments having a preconfigured length, and wherein the computer generates a biometric embedding and a spoofprint embedding for each segment.
21. The system according to claim 12 , wherein the computer is further configured to generate a lip-sync score by applying a lip sync estimation engine of the machine-learning architecture on the audiovisual data, and wherein the computer generates the final output score using the lip-sync score.
22. The system of claim 12 , wherein the audiovisual data contains audio data, image data, or both audio data and image data.