Speech2Face : Learning the face behind a voice
The researchers from MIT CSAIL published this research paper with a reason that how much face information can be obtained from speech.
Link here: https://speech2face.github.io/
Abstract:
“Whenever we hear a voice of a person, our brain expects the gender, age and imagines how the person might look based on the voice. The idea of building an AI system which is capable of identifying the gender, age, ethnicity etc., and constructing basic identifiable facial image of a person with neutral expression simply based on their voice. This task can be achieved through a deep neural network. We design and train a model using thousands of videos available in the internet. The model learns the voice-face correlations during training which later allows the model to build an image. This can be useful for security forces in identifying criminals using their speech.”
Working model
