Robust and Unbounded Length Generalization in Autoregressive Transformer-Based Text-to-Speech

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Battenberg, Eric, Skerry-Ryan, RJ, Stanton, Daisy, Mariooryad, Soroosh, Shannon, Matt, Salazar, Julian, Kao, David
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916650151837696
author Battenberg, Eric
Skerry-Ryan, RJ
Stanton, Daisy
Mariooryad, Soroosh
Shannon, Matt
Salazar, Julian
Kao, David
author_facet Battenberg, Eric
Skerry-Ryan, RJ
Stanton, Daisy
Mariooryad, Soroosh
Shannon, Matt
Salazar, Julian
Kao, David
contents Autoregressive (AR) Transformer-based sequence models are known to have difficulty generalizing to sequences longer than those seen during training. When applied to text-to-speech (TTS), these models tend to drop or repeat words or produce erratic output, especially for longer utterances. In this paper, we introduce enhancements aimed at AR Transformer-based encoder-decoder TTS systems that address these robustness and length generalization issues. Our approach uses an alignment mechanism to provide cross-attention operations with relative location information. The associated alignment position is learned as a latent property of the model via backpropagation and requires no external alignment information during training. While the approach is tailored to the monotonic nature of TTS input-output alignment, it is still able to benefit from the flexible modeling power of interleaved multi-head self- and cross-attention operations. A system incorporating these improvements, which we call Very Attentive Tacotron, matches the naturalness and expressiveness of a baseline T5-based TTS system, while eliminating problems with repeated or dropped words and enabling generalization to any practical utterance length.
format Preprint
id arxiv_https___arxiv_org_abs_2410_22179
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Robust and Unbounded Length Generalization in Autoregressive Transformer-Based Text-to-Speech
Battenberg, Eric
Skerry-Ryan, RJ
Stanton, Daisy
Mariooryad, Soroosh
Shannon, Matt
Salazar, Julian
Kao, David
Computation and Language
Machine Learning
Sound
Audio and Speech Processing
Autoregressive (AR) Transformer-based sequence models are known to have difficulty generalizing to sequences longer than those seen during training. When applied to text-to-speech (TTS), these models tend to drop or repeat words or produce erratic output, especially for longer utterances. In this paper, we introduce enhancements aimed at AR Transformer-based encoder-decoder TTS systems that address these robustness and length generalization issues. Our approach uses an alignment mechanism to provide cross-attention operations with relative location information. The associated alignment position is learned as a latent property of the model via backpropagation and requires no external alignment information during training. While the approach is tailored to the monotonic nature of TTS input-output alignment, it is still able to benefit from the flexible modeling power of interleaved multi-head self- and cross-attention operations. A system incorporating these improvements, which we call Very Attentive Tacotron, matches the naturalness and expressiveness of a baseline T5-based TTS system, while eliminating problems with repeated or dropped words and enabling generalization to any practical utterance length.
title Robust and Unbounded Length Generalization in Autoregressive Transformer-Based Text-to-Speech
topic Computation and Language
Machine Learning
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2410.22179