DiTAR: Diffusion Transformer Autoregressive Modeling for Speech Generation

TL;DR

DiTAR introduces a novel patch-based autoregressive framework combining diffusion transformers and language models, significantly improving speech generation quality and efficiency with state-of-the-art results in robustness, speaker similarity, and naturalness.

Contribution

The paper presents DiTAR, a new diffusion transformer autoregressive model that enhances continuous speech generation by reducing computational load and improving scalability.

Findings

Achieves state-of-the-art zero-shot speech generation performance.

Demonstrates superior scalability in extensive analysis.

Balances diversity and determinism through temperature control during inference.

Abstract

Several recent studies have attempted to autoregressively generate continuous speech representations without discrete speech tokens by combining diffusion and autoregressive models, yet they often face challenges with excessive computational loads or suboptimal outcomes. In this work, we propose Diffusion Transformer Autoregressive Modeling (DiTAR), a patch-based autoregressive framework combining a language model with a diffusion transformer. This approach significantly enhances the efficacy of autoregressive models for continuous tokens and reduces computational demands. DiTAR utilizes a divide-and-conquer strategy for patch generation, where the language model processes aggregated patch embeddings and the diffusion transformer subsequently generates the next patch based on the output of the language model. For inference, we propose defining temperature as the time point of introducing noise during the reverse diffusion ODE to balance diversity and determinism. We also show in the extensive scaling analysis that DiTAR has superb scalability. In zero-shot speech generation, DiTAR achieves state-of-the-art performance in robustness, speaker similarity, and naturalness.