Abstract: System and method detecting synthetic speech may include, using a processor: training a prosody extractor by: providing a training speech sample to an encoder-decoder (codec) model to generate a channel degraded speech sample; providing the channel degraded speech sample to a prosody extractor to extract a prosody embedding; providing to a speech synthesis model the prosody embedding, a codec embedding representing the codec model, speaker identity information and a text representation to generate a spectrogram of the training speech sample; and training the speech synthesis model and the prosody extractor using a loss function defined on the spectrogram generated by the speech synthesis model compared with a spectrogram of the channel degraded speech sample; and using the trained prosody extractor to detect synthetic speech.

Abstract: System and method for matching a voice sample to a facial image may include, using a processor: obtaining a voice sample and a facial image; calculating a plurality of voice metaproperties from the voice sample; calculating a plurality of image metaproperties from image sample, where each of the image metaproperties corresponds to one of the voice metaproperties; and determining a level of match between the voice sample and the facial image, based on the plurality of voice metaproperties and the plurality of image metaproperties.

Abstract: System and method for reconstructing a facial image of a speaker from a voice sample of the speaker including: selecting, from a dataset associating each of a plurality of voice samples to a particular facial image, a subset of voice samples that are similar to the voice sample of the speaker, and reconstructing the facial image of the speaker by unifying the faces associated with each of the voice samples in the subset of voice samples.

Abstract: System and method for voice-face matching, including: obtaining a voice sample, a first facial image of a first person, and a second facial image of a second person; calculating a distance between the first facial image and the second facial image; using a voice-face matching model, calculating a latent space vector for each of the voice sample and the first and second facial images; and training the voice-face matching model using a loss function that decreases a distance between the latent space vector of the voice sample of the first person and the latent space vector of the first facial image, and increases a distance between the latent space vector of the voice sample of the first person and the latent space vector of the second facial image, wherein the loss function is adjusted based on the distance between the first facial image and the second facial image.