Spoofing detection apparatus, spoofing detection method, and computer-readable storage medium

1. A spoofing detection apparatus comprising:

one or more processors; and

a memory storing instructions executable by the one or more processors to:

extract different types of spectrograms, including FFT spectrograms and CQT spectrograms, from speech data, and integrate the different types of spectrograms to create a multi-channel 3D spectrogram by fusing different types of spectrograms into the created multi-channel 3D spectrogram; and

evaluate the created multi-channel 3D spectrogram by applying the created multi-channel 3D spectrogram to a classifier constructed using labeled multi-channel spectrograms as training data and classify the created multi-channel 3D spectrogram as either genuine or spoof, wherein

the CQT spectrograms have additional zero elements placed therein, and the FFT spectrograms are zero-padded, so as to have a dimension in frequency equal to a same designated number that is equal to the dimension in frequency of either the CQT spectrograms as extracted or the FFT spectrograms as extracted.

2. The spoofing detection apparatus according to claim 1 , wherein

the created multi-channel 3D spectrogram and a label corresponding to the speech data are used as the training data to construct the classifier.

3. The spoofing detection apparatus according to claim 1 ,

wherein the different type of spectrograms are integrated by stacking.

4. The spoofing detection apparatus according to claim 1 ,

wherein the different type of spectrograms are integrated by concatenation.

5. The spoofing detection apparatus according to claim 1 ,

wherein the different types of spectrograms are resampled into a same size before the multi-channel 3D spectrogram is created.

6. The spoofing detection apparatus according to claim 1 ,

wherein the different types of spectrograms are zero-padded into a same size before the multi-channel 3D spectrogram is created.

7. A spoofing detection method comprising:

extracting, by a processor, different types of spectrograms, including FFT spectrograms and CQT spectrograms, from speech data, and integrating the different types of spectrograms to create a multi-channel 3D spectrogram by fusing different types of spectrograms into the created multi-channel 3D spectrogram; and

evaluating, by the processor, the created multi-channel 3D spectrogram by applying the created multi-channel 3D spectrogram to a classifier constructed using labeled multi-channel spectrograms as training data and classifying the created multi-channel 3D spectrogram as either genuine or spoof, wherein

the CQT spectrograms have additional zero elements placed therein, and the FFT spectrograms are zero-padded, so as to have a dimension in frequency equal to a same designated number that is equal to the dimension in frequency of either the CQT spectrograms as extracted or the FFT spectrograms as extracted.

8. A non-transitory computer-readable storage medium storing a program that includes commands for causing a computer to execute:

extracting different types of spectrograms, including FFT spectrograms and CQT spectrograms, from speech data, and integrating the different types of spectrograms to create a multi-channel 3D spectrogram by fusing different types of spectrograms into the created multi-channel 3D spectrogram; and

evaluating the created multi-channel 3D spectrogram by applying the created multi-channel 3D spectrogram to a classifier constructed using labeled multi-channel spectrograms as training data and classifying the created multi-channel 3D spectrogram as either genuine or spoof, wherein

the CQT spectrograms have additional zero elements placed therein, and the FFT spectrograms are zero-padded, so as to have a dimension in frequency equal to a same designated number that is equal to the dimension in frequency of either the CQT spectrograms as extracted or the FFT spectrograms as extracted.

9. The spoofing detection method according to claim 7 , wherein

the created multi-channel 3D spectrogram and a label corresponding to the speech data are used as the training data to construct the classifier.

10. The spoofing detection method according to claim 7 ,

wherein the different type of spectrograms are integrated by stacking.

11. The spoofing detection method according to claim 7 ,

wherein the different type of spectrograms are integrated by concatenation.

12. The spoofing detection method according to claim 7 ,

wherein the different types of spectrograms are resampled into a same size before the multi-channel 3D spectrogram is created.

13. The spoofing detection method according to claim 7 ,

wherein the different types of spectrograms are zero-padded into a same size before the multi-channel 3D spectrogram is created.

14. The non-transitory computer-readable storage medium according to claim 8 ,

wherein the created multi-channel 3D spectrogram and a label corresponding to the speech data are used as the training data to construct the classifier.

15. The non-transitory computer-readable storage medium according to claim 8 ,

wherein, the different type of spectrograms are integrated by stacking-.

16. The non-transitory computer-readable storage medium according to claim 8 ,

wherein, the different type of spectrograms are integrated by concatenation.

17. The non-transitory computer-readable storage medium according to claim 8 ,

wherein, the different types of spectrograms are resampled into a same size before the multi-channel 3D spectrogram is created.

18. The non-transitory computer-readable storage medium according to claim 8 ,

wherein the different types of spectrograms are zero-padded into a same size before creating the multi-channel 3D spectrogram is created.