Dual-Space Constrained Face-Based Zero-Shot Text-to-Speech Synthesis

Jianrong Wang1, Shengjie Zhou1, Ju Zhang2,*, Dengcheng Hu1, Qi Li3
1College of Intelligence and Computing, Tianjin University
2Technical College for the Deaf, Tianjin University of Technology
3School of Electrical and Information Engineering, Tianjin University
Interspeech 2026

*Corresponding author
Overview of the proposed DSC-TTS framework

Overview of the proposed DSC-TTS framework.

Abstract

A human face conveys rich cues about speaker identity, enabling face-based zero-shot text-to-speech (TTS) for unseen speakers. However, in modular face-based TTS systems, the acoustic model is typically trained on speech-derived embeddings, while face-derived representations are introduced only at inference time, often resulting in identity drift. We propose Dual-Space Constrained TTS (DSC-TTS), a modular framework that enforces identity consistency during acoustic model training in both the speaker embedding space and a shared identity space learned through face-voice alignment. By constraining representations across these complementary spaces, the proposed framework improves speaker identity stability while preserving speech quality. Experiments demonstrate higher speaker similarity and stronger identity consistency than existing face-based TTS methods.

Audio Samples Comparison

For modular baselines (Face2Speech, SYNTHE-SEES, Face-StyleSpeech), we adopt their original face-voice alignment architectures within a unified TTS framework. All systems share the same acoustic model and speaker encoder, with differences in alignment module design and speaker embedding learning.

VoxCeleb2 Dataset

Face images from VoxCeleb2 and text sentences from LibriTTS test-clean set are used for synthesis.

Face Text Ground Truth FaceTTS Face2Speech SYNTHE-SEES Face-StyleSpeech DSC-TTS (Ours)

LRS2 Dataset

Sample-level evaluation on LRS2 tri-modal dataset under corpus mismatch.

Face Text Ground Truth FaceTTS Face2Speech SYNTHE-SEES Face-StyleSpeech DSC-TTS (Ours)

BibTeX