arXiv Machine Learning

Attention-Based Adaptive Policies for Simultaneous Speech-to-Text Translation

The paper introduces two new policies—Recent Frame Attention Policy (RFAP) and Dual-Condition Attention Policy (DCAP)—for simultaneous speech-to-text translation. These policies leverage the cross‑attention mechanism of encoder‑decoder models to determine optimal moments for partial translation, enabling offline models to operate in streaming scenarios without extra training. Experiments on the CVSS‑C corpus show RFAP improves BLEU scores by up to 4.0 points while cutting delay by nearly one second, and DCAP maintains high quality at very low latency.

arXiv Computation and Language
Sep 11

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

The paper introduces Hikari, a policy‑free, end‑to‑end model that performs simultaneous speech‑to‑text translation and streaming transcription. It employs a Decoder Time Dilation mechanism to mitigate overuse of WAIT tokens during training and a supervised fine‑tuning strategy that helps the model recover from delays, improving the quality‑latency trade‑off. Despite its modest size, Hikari achieves competitive translation quality at consistently low latency, outperforming larger published IWSLT 2026 submissions and proprietary API systems on en‑ja, en‑de, and en‑ru tasks.

By Roman Koshkin, Jeon Haesung, Lianbo Liu, Hao Shi, Mengjie Zhao, Yusuke Fujita, Yui Sudo
arXiv Computation and Language
Nov 5

CRESiST: Self-Enhancing Explorative Policy Learning for Simultaneous Speech Translation

CRESiST introduces an end‑to‑end interleaved read‑write framework that leverages an omni‑modal LLM for simultaneous speech translation. By interleaving speech chunks with text tokens, the system fully reuses the KV cache and supports segmentation‑free, infinite‑length inference, eliminating the need for external voice activity detection. An Actor‑Critic self‑enhancement pipeline trains an RNN‑T actor to sample read‑write trajectories, while an LLM‑based critic evaluates them using semantic and latency rewards, distilling the optimal streaming policy back to the actor. Experiments on sentence‑ and document‑level benchmarks show that CRESiST sets new state‑of‑the‑art quality‑latency Pareto frontiers, outperforming existing cascaded and static‑alignment baselines.

By Chenyang Le, Pei Zhang, Yuxiang Chen, Bing Han, Derek F. Wong, Baosong Yang, Yanmin Qian
arXiv AI
Sep 4

X-Translator: A Real-Time Multilingual Speaker-Aware Speech-to-Speech Translation System

X-Translator is a low‑cost, modular real‑time speech‑to‑speech translation system that integrates streaming ASR, machine translation, and prompt‑conditioned TTS, managed by a session‑level runtime controller. It uses incremental segment commitment to stabilize ASR streams and an online speaker prompt manager to maintain speaker consistency across multi‑speaker conversations. The system is evaluated on translation quality, speech naturalness, latency, and speaker preservation using OpenSTBench, and its code and demo are publicly available on GitHub.

By Yuxiang Zhao, Yichi Zhang, Yanjie An, Yanqiao Zhu, Zhanxun Liu, Yushen Chen, Qixi Zheng, Haina Zhu, Yunchong Xiao, Keqi Deng, Shuai Fan, Kai Yu, Xie Chen
arXiv Computation and Language
Oct 26

DIAL-S2ST: Learning Latency Control for Simultaneous Speech-to-Speech Translation

DIAL‑S2ST is a post‑training method that learns to control inference‑time latency in simultaneous speech‑to‑speech translation models. It reuses pretrained translation capabilities through self‑distillation from lookahead predictions and limits large text distillation losses to address the imitation gap, providing supervision for earlier emission without paired target data or reward computation. By jointly training a single LoRA adapter across scales with matched lookahead guidance, the learned timing can be adjusted at inference, and experiments on French‑to‑English translation with Hibiki show that only 10 hours of source speech enable this adaptation, yielding flexible latency control with a favorable quality‑latency trade‑off.

By Jeongsoo Choi, Junseok Ahn, Joon Son Chung
arXiv Computation and Language
Sep 28

All In Good Time: Causality-Aware Framework for LLM-Based Simultaneous Speech-to-Speech Translation

The paper introduces FAST-CAP, a causality‑aware framework for simultaneous speech‑to‑speech translation that combines a factorized S2ST architecture, an adaptive policy, and a new latency metric. It employs a novel data pipeline to generate high‑fidelity, causally aligned segments, improving voice transfer and reducing the need for large training datasets. Experiments on Spanish, German, and French demonstrate that FAST‑CAP outperforms fixed‑policy baselines, achieving up to +1.2 BLEU, 26% lower latency, and a 38.8% relative latency reduction while maintaining speaker fidelity.

By Amir Hussein, Enas Albasiri, Travis M. Bartley, Nourchene Ferchichi, Ke Hu, Harishchandra Dubey, Myungjong Kim, Zhehuai Chen, Oluwatobi Olabiyi, Sanjeev Khudanpur
arXiv Computation and Language
Sep 21

Dynamic Lagging using Stable-Prefix Training for Simultaneous Translation

The paper introduces a training strategy for cascaded simultaneous speech translation that allows the system to dynamically decide how much of the source prefix to translate. By fine‑tuning a large language model (Qwen3‑8B) on stable prefixes—pairs of source prefixes and the longest shared translation with the full sentence—the authors enable contextual read‑write decisions beyond fixed wait‑k or target‑suffix deletion. Experiments on English‑to‑German, Japanese, and Chinese demonstrate that stable prefixes improve the quality‑latency tradeoff across various test sets.

By Hieu Hoang, Amittai Axelrod, Matt Post