RetrieverTTS: Modeling Decomposed Factors for Text-Based Speech Insertion

TL;DR

RetrieverTTS introduces a decompose-and-edit paradigm for text-based speech insertion that separates global timbre/style factors from local text and prosody, enabling arbitrary-length insertions and full sentence generation. By extracting global-factor tokens via cross-attention and injecting them through link-attention, plus a prosody smoothing training task and adversarial training, the method achieves high speaker similarity, natural prosody, and realistic voice quality. The approach is evaluated on LibriTTS with MOS-based subjective tests and ablations, showing state-of-the-art performance for naturalness and similarity and robustness across insertion lengths. The work provides a practical, zero-shot capable framework for high-quality text-based speech insertion with broad applicability in video and multimedia editing.

Abstract

This paper proposes a new "decompose-and-edit" paradigm for the text-based speech insertion task that facilitates arbitrary-length speech insertion and even full sentence generation. In the proposed paradigm, global and local factors in speech are explicitly decomposed and separately manipulated to achieve high speaker similarity and continuous prosody. Specifically, we proposed to represent the global factors by multiple tokens, which are extracted by cross-attention operation and then injected back by link-attention operation. Due to the rich representation of global factors, we manage to achieve high speaker similarity in a zero-shot manner. In addition, we introduce a prosody smoothing task to make the local prosody factor context-aware and therefore achieve satisfactory prosody continuity. We further achieve high voice quality with an adversarial training stage. In the subjective test, our method achieves state-of-the-art performance in both naturalness and similarity. Audio samples can be found at https://ydcustc.github.io/retrieverTTS-demo/.

Figures (1)

• Figure 1: Model architecture.