Attention-Constrained Inference for Robust Decoder-Only Text-to-Speech

TL;DR

This paper introduces Attention-Constrained Inference (ACI), a novel method that leverages attention maps to improve the monotonicity and accuracy of decoder-only text-to-speech models without retraining.

Contribution

The paper proposes a new inference technique using attention map analysis to enhance speech synthesis quality in decoder-only TTS models.

Findings

WER reduced by up to 20.5% with ACI

Maintains naturalness and speaker similarity

Identifies attention heads indicating speech-text alignment

Abstract

Recent popular decoder-only text-to-speech models are known for their ability of generating natural-sounding speech. However, such models sometimes suffer from word skipping and repeating due to the lack of explicit monotonic alignment constraints. In this paper, we notice from the attention maps that some particular attention heads of the decoder-only model indicate the alignments between speech and text. We call the attention maps of those heads Alignment-Emerged Attention Maps (AEAMs). Based on this discovery, we propose a novel inference method without altering the training process, named Attention-Constrained Inference (ACI), to facilitate monotonic synthesis. It first identifies AEAMs using the Attention Sweeping algorithm and then applies constraining masks on AEAMs. Our experimental results on decoder-only TTS model VALL-E show that the WER of synthesized speech is reduced by up to 20.5% relatively with ACI while the naturalness and speaker similarity are comparable.