RobustL2S: Speaker-Specific Lip-to-Speech Synthesis exploiting Self-Supervised Representations

TL;DR

RobustL2S introduces a modular Lip-to-Speech synthesis framework that leverages self-supervised visual features to improve disentanglement of speech content and speaker traits, achieving state-of-the-art results.

Contribution

The paper presents a novel modular approach using self-supervised features for better speech content disentanglement in Lip-to-Speech synthesis.

Findings

Achieves state-of-the-art performance on multiple datasets

Effectively disentangles speech content from speaker traits

Outperforms existing non-autoregressive models

Abstract

Significant progress has been made in speaker dependent Lip-to-Speech synthesis, which aims to generate speech from silent videos of talking faces. Current state-of-the-art approaches primarily employ non-autoregressive sequence-to-sequence architectures to directly predict mel-spectrograms or audio waveforms from lip representations. We hypothesize that the direct mel-prediction hampers training/model efficiency due to the entanglement of speech content with ambient information and speaker characteristics. To this end, we propose RobustL2S, a modularized framework for Lip-to-Speech synthesis. First, a non-autoregressive sequence-to-sequence model maps self-supervised visual features to a representation of disentangled speech content. A vocoder then converts the speech features into raw waveforms. Extensive evaluations confirm the effectiveness of our setup, achieving state-of-the-art performance on the unconstrained Lip2Wav dataset and the constrained GRID and TCD-TIMIT datasets. Speech samples from RobustL2S can be found at https://neha-sherin.github.io/RobustL2S/