Salvato in:
Dettagli Bibliografici
Autori principali: Koshkin, Roman, Haesung, Jeon, Liu, Lianbo, Shi, Hao, Zhao, Mengjie, Fujita, Yusuke, Sudo, Yui
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:https://arxiv.org/abs/2603.11578
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915856737370112
author Koshkin, Roman
Haesung, Jeon
Liu, Lianbo
Shi, Hao
Zhao, Mengjie
Fujita, Yusuke
Sudo, Yui
author_facet Koshkin, Roman
Haesung, Jeon
Liu, Lianbo
Shi, Hao
Zhao, Mengjie
Fujita, Yusuke
Sudo, Yui
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2603_11578
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Streaming Translation and Transcription Through Speech-to-Text Causal Alignment
Koshkin, Roman
Haesung, Jeon
Liu, Lianbo
Shi, Hao
Zhao, Mengjie
Fujita, Yusuke
Sudo, Yui
Computation and Language
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.
title Streaming Translation and Transcription Through Speech-to-Text Causal Alignment
topic Computation and Language
url https://arxiv.org/abs/2603.11578