Streaming Translation and Transcription Through Speech-to-Text Causal Alignment

TL;DR

Hikari is an end-to-end, policy-free streaming translation model that uses probabilistic WAIT tokens and decoder time dilation to improve translation quality and latency trade-offs across multiple languages.

Contribution

The paper introduces Hikari, a novel end-to-end model for streaming speech translation that eliminates reliance on heuristics and improves performance with new training strategies.

Findings

Achieves state-of-the-art BLEU scores in multiple languages.

Effectively balances quality and latency in streaming translation.

Outperforms recent baselines in low- and high-latency regimes.

Abstract

Simultaneous machine translation (SiMT) has traditionally relied on offline machine translation models coupled with human-engineered heuristics or learned policies. We propose Hikari, a policy-free, fully end-to-end model that performs simultaneous speech-to-text translation and streaming transcription by encoding READ/WRITE decisions into a probabilistic WAIT token mechanism. We also introduce Decoder Time Dilation, a mechanism that reduces autoregressive overhead and ensures a balanced training distribution. Additionally, we present a supervised fine-tuning strategy that trains the model to recover from delays, significantly improving the quality-latency trade-off. Evaluated on English-to-Japanese, German, and Russian, Hikari achieves new state-of-the-art BLEU scores in both low- and high-latency regimes, outperforming recent baselines.