Speech-driven animation using one or more neural networks
Apparatuses, systems, and techniques are presented to generate digital content. In at least one embodiment, one or more neural networks are used to generate video information based at least in part upon voice information and a combination of image features and facial landmarks corresponding to one or more images of a person.
1. A processor, comprising:
one or more circuits to;
predict facial landmarks of a person uttering a speech based, at least in part, upon voice information of the speech and one or more images of the person;
predict keypoints based, at least in part, upon the predicted facial landmarks and the voice information; and
generate, using one or more neural networks, video information of the person uttering the speech based, at least in part, upon the predicted keypoints.
2. The processor of claim 1 , wherein the one or more circuits are further to use a first neural network to determine a set of mouth landmarks and a second neural network to determine a set of face landmarks for the person corresponding to the voice information, wherein the predicted facial landmarks include both the set of face landmarks and the set of mouth landmarks.
3. The processor of claim 2 , wherein the one or more circuits are further to use a keypoint determination neural network to predict the keypoints for the person corresponding to the voice information based, at least in part, upon the predicted facial landmarks.
4. The processor of claim 3 , wherein the one or more circuits are further to use a generative neural network to generate the video information based, at least in part, upon the voice information and the predicted keypoints, and further upon pose information predicted by a pose generation network.
5. The processor of claim 1 , wherein the one or more circuits are further to generate a set of frontalized landmarks normalized for a face and mouth of the person to provide as input, along with a feature representation of the voice information, to at least one landmark prediction neural network.
6. The processor of claim 1 , wherein the one or more circuits are further to use at least one emotion prediction module to predict one or more emotional states for individual frames of the video information and generate the video information to reflect the predicted emotional states.
7. A system, comprising:
one or more processors to:
predict facial landmarks of a person uttering a speech based, at least in part, upon voice information of the speech and one or more images of the person;
predict keypoints based, at least in part, upon the predicted facial landmarks and the voice information; and
generate, using one or more neural networks, video information of the person uttering the speech based, at least in part, upon the predicted keypoints.
8. The system of claim 7 , wherein the one or more processors are further to use a first neural network to determine a set of mouth landmarks and a second neural network to determine a set of face landmarks for the person corresponding to the voice information, wherein the predicted facial landmarks include both the set of face landmarks and the set of mouth landmarks.
9. The system of claim 8 , wherein the one or more processors are further to use a keypoint determination neural network to predict the keypoints for the person corresponding to the voice information based, at least in part, upon the predicted facial landmarks.
10. The system of claim 9 , wherein the one or more processors are further to use a generative neural network to generate the video information based, at least in part, upon the voice information and the predicted keypoints, and further upon pose information predicted by a pose generation network.
11. The system of claim 7 , wherein the one or more processors are further to generate a set of frontalized landmarks normalized for a face and mouth of the person to provide as input, along with a feature representation of the voice information, to at least one landmark prediction neural network.
12. The system of claim 7 , wherein the one or more processors are further to use at least one emotion prediction module to predict one or more emotional states for individual frames of the video information and generate the video information to reflect the predicted emotional states.
13. A method, comprising:
predicting facial landmarks of a person uttering a speech based, at least in part, upon voice information of the speech and one or more images of the person;
predicting keypoints based, at least in part, upon the predicted facial landmarks and the voice information; and
generating, using one or more neural networks, video information of the person uttering the speech based, at least in part, upon the predicted keypoints.
14. The method of claim 13 , further comprising: using a first neural network to determine a set of mouth landmarks and a second neural network to determine a set of face landmarks for the person corresponding to the voice information, wherein the predicted facial landmarks include both the set of face landmarks and the set of mouth landmarks.
15. The method of claim 14 , further comprising: using a keypoint determination neural network to predict the keypoints for the person corresponding to the voice information based, at least in part, upon the predicted facial landmarks.
16. The method of claim 15 , further comprising: using a generative neural network to generate the video information based, at least in part, upon the voice information and the predicted keypoints, and further upon pose information predicted by a pose generation network.
17. The method of claim 13 , further comprising:
generating a set of frontalized landmarks normalized for a face and mouth of the person to provide as input, along with a feature representation of the voice information, to at least one landmark prediction neural network.
18. The method of claim 13 , further comprising:
using at least one emotion prediction module to predict one or more emotional states for individual frames of the video information and generate the video information to reflect the predicted emotional states.
19. A video generation system, comprising:
one or more processors to:
predict facial landmarks of a person uttering a speech based, at least in part, upon voice information of the speech and one or more images of the person;
predict keypoints based, at least in part, upon the predicted facial landmarks and the voice information; and
generate, using one or more neural networks, video information of the person uttering the speech based, at least in part, upon the predicted keypoints; and
memory for storing network parameters for the one or more first neural networks.
20. The video generation system of claim 19 , wherein the one or more processors are further to use a first neural network to determine a set of mouth landmarks and a second neural network to determine a set of face landmarks for the person corresponding to the voice information, wherein the predicted facial landmarks include both the set of face landmarks and the set of mouth landmarks.
21. The video generation system of claim 20 , wherein the one or more processors are further to use a keypoint determination neural network to predict the keypoints for the person corresponding to the voice information based, at least in part, upon the predicted facial landmarks.
22. The video generation system of claim 21 , wherein the one or more processors are further to use a generative neural network to generate the video information based, at least in part, upon the voice information and the predicted keypoints, and further upon pose information predicted by a pose generation network.
23. The video generation system of claim 19 , wherein the one or more processors are further to generate a set of frontalized landmarks normalized for a face and mouth of the person to provide as input, along with a feature representation of the voice information, to at least one landmark prediction neural network.
24. The video generation system of claim 19 , wherein the one or more processors are further to use at least one emotion prediction module to predict one or more emotional states for individual frames of the video information and generate the video information to reflect the predicted emotional states.