MoE-TTS: Enhancing Out-of-Domain Text Understanding for Description-based TTS via Mixture-of-Experts

TL;DR

MoE-TTS introduces a mixture-of-experts approach to improve out-of-domain text understanding in description-based TTS, leveraging pre-trained language models with specialized speech modality experts for more accurate speech synthesis.

Contribution

The paper proposes MoE-TTS, a novel mixture-of-experts model that enhances out-of-domain text understanding in TTS by augmenting pre-trained LLMs with speech-specific experts while keeping the core LLM frozen.

Findings

MoE-TTS outperforms existing models on out-of-domain description tests.

Commercial TTS systems struggle with carefully designed out-of-domain inputs.

MoE-TTS generates speech that more accurately reflects diverse descriptions.

Abstract

Description-based text-to-speech (TTS) models exhibit strong performance on in-domain text descriptions, i.e., those encountered during training. However, in real-world applications, the diverse range of user-generated descriptions inevitably introduces numerous out-of-domain inputs that challenge the text understanding capabilities of these systems. To address this issue, we propose MoE-TTS, a description-based TTS model designed to enhance the understanding of out-of-domain text descriptions. MoE-TTS employs a modality-based mixture-of-experts (MoE) approach to augment a pre-trained textual large language model (LLM) with a set of specialized weights adapted to the speech modality while maintaining the original LLM frozen during training. This approach allows MoE-TTS to effectively leverage the pre-trained knowledge and text understanding abilities of textual LLMs. Our experimental results indicate that: first, even the most advanced closed-source commercial products can be challenged by carefully designed out-of-domain description test sets; second, MoE-TTS achieves superior performance in generating speech that more accurately reflects the descriptions. We encourage readers to listen to the demos at https://welkinyang.github.io/MoE-TTS/.