Speech Prediction in Silent Videos using Variational Autoencoders
This work addresses the challenge of predicting speech in silent videos for applications like CGI, video editing, and assisting people with impairments, but it is incremental as it builds on existing methods by incorporating stochasticity.
The paper tackles the problem of generating speech from silent videos by addressing the multimodal nature of audio-visual relationships, using a stochastic model that combines recurrent neural networks and variational autoencoders to learn conditional distributions, and demonstrates performance on the GRID dataset with standard benchmarks.
Understanding the relationship between the auditory and visual signals is crucial for many different applications ranging from computer-generated imagery (CGI) and video editing automation to assisting people with hearing or visual impairments. However, this is challenging since the distribution of both audio and visual modality is inherently multimodal. Therefore, most of the existing methods ignore the multimodal aspect and assume that there only exists a deterministic one-to-one mapping between the two modalities. It can lead to low-quality predictions as the model collapses to optimizing the average behavior rather than learning the full data distributions. In this paper, we present a stochastic model for generating speech in a silent video. The proposed model combines recurrent neural networks and variational deep generative models to learn the auditory signal's conditional distribution given the visual signal. We demonstrate the performance of our model on the GRID dataset based on standard benchmarks.