Efficient Monotonic Multihead Attention

TL;DR

This paper presents EMMA, a novel monotonic multihead attention model for simultaneous speech-to-text translation, featuring stable alignment estimation, improved training strategies, and achieving state-of-the-art results on Spanish and English tasks.

Contribution

The paper introduces EMMA, a new efficient monotonic multihead attention mechanism with unbiased alignment estimation and enhanced training methods for improved translation performance.

Findings

Achieves state-of-the-art results in speech-to-text translation

Demonstrates stable and unbiased alignment estimation

Improves training and inference efficiency

Abstract

We introduce the Efficient Monotonic Multihead Attention (EMMA), a state-of-the-art simultaneous translation model with numerically-stable and unbiased monotonic alignment estimation. In addition, we present improved training and inference strategies, including simultaneous fine-tuning from an offline translation model and reduction of monotonic alignment variance. The experimental results demonstrate that the proposed model attains state-of-the-art performance in simultaneous speech-to-text translation on the Spanish and English translation task.