Small-E: Small Language Model with Linear Attention for Efficient Speech Synthesis

TL;DR

Small-E introduces a recurrent architecture with linear attention for efficient, high-quality speech synthesis, overcoming transformer limitations in sequence length and alignment biases, enabling better zero-shot voice cloning.

Contribution

The paper proposes a novel recurrent architecture with linear attention and specialized cross-attention for efficient long-sequence speech synthesis, improving zero-shot voice cloning.

Findings

Achieves state-of-the-art zero-shot voice cloning performance.

Efficient training on long speech samples.

Reduces issues of repeating and skipping in speech generation.

Abstract

Recent advancements in text-to-speech (TTS) powered by language models have showcased remarkable capabilities in achieving naturalness and zero-shot voice cloning. Notably, the decoder-only transformer is the prominent architecture in this domain. However, transformers face challenges stemming from their quadratic complexity in sequence length, impeding training on lengthy sequences and resource-constrained hardware. Moreover they lack specific inductive bias with regards to the monotonic nature of TTS alignments. In response, we propose to replace transformers with emerging recurrent architectures and introduce specialized cross-attention mechanisms for reducing repeating and skipping issues. Consequently our architecture can be efficiently trained on long samples and achieve state-of-the-art zero-shot voice cloning against baselines of comparable size. Our implementation and demos are available at https://github.com/theodorblackbird/lina-speech.