FARV: Leveraging Facial and Acoustic Representation in Vocoder For Video-to-Speech Synthesis
Keywords: Video-to-speech (V2S), vocoder, speech synthesis
TL;DR: The paper proposes a multi-modal unit-based vocoder designed for V2S, which is able to sustain V2S domain gap and preserve speaker characteristics at the same time.
Abstract: In this paper, we introduce FARV, a vocoder specifically designed for Video-to-Speech (V2S) synthesis, which integrates both facial embeddings and acoustic units to generate speech waveforms. By sharing the acoustic unit vocabulary in our two-stage V2S pipeline, FARV effectively bridges the domain gap between the visual frontend and the vocoder without requiring finetuning. Furthermore, by embedding visual speaker images into the acoustic unit representations, FARV enhances its ability to preserve speaker identity. Experimental results demonstrate that FARV achieves leading scores in intelligibility and strikes a favorable balance between speaker characterisitcs preservation and acoustic quality, making it well-suited for practical V2S applications.
Supplementary Material: zip
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3349
Loading