MFLA: Monotonic Finite Look-ahead Attention for Streaming Speech Recognition

2025-06-04 · Yinfeng Xia, Huiyan Li, Chenyang Le, Manhong Wang, Yutao Sun, Xingyang Ma, Yanmin Qian

Applying large pre-trained speech models like Whisper has shown promise in reducing training costs for various speech tasks. However, integrating these models into streaming systems remains a challenge. This paper presents a novel prefix-to-prefix training framework for streaming recognition by fine-tuning the Whisper. We introduce the Continuous Integrate-and-Fire mechanism to establish a quasi-monotonic alignment between continuous speech sequences and discrete text tokens. Additionally, we design Monotonic Finite Look-ahead Attention, allowing each token to attend to infinite left-context and finite right-context from the speech sequences. We also employ the wait-k decoding strategy to simplify the decoding process while ensuring consistency between training and testing. Our theoretical analysis and experiments demonstrate that this approach achieves a controllable trade-off between latency and quality, making it suitable for various streaming applications.

📄 PDF Abstract BibTeX arXiv:2506.03722

Code (0)

등록된 구현이 없습니다.

Tasks

speech-recognition Speech Recognition

Methods 이 논문이 사용한 방법론

Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…

Attention 설명 없음

Monotonic Infinite Lookback Attention for Simultaneous Machine Translation

2019-06-12 · ACL 2019 7 · Naveen Arivazhagan, Colin Cherry, Wolfgang Macherey, Chung-Cheng Chiu 외

Simultaneous machine translation begins to translate each source sentence before the source speaker is finished speaking, with applications to live and streaming scenarios. Simultaneous systems must carefully schedule th…

Machine Translation NMT Sentence Translation

Streaming T5-based Text-to-Speech Synthesis with Limited Lookahead

2026-06-20 · Muyang Du, Jason Roche, Junjie Lai arxiv

Streaming text-to-speech synthesis in cascaded LLM-TTS systems still faces latency challenges as most TTS models require full context before initiating generation. We present S5-TTS, a streaming variant of T5-TTS that en…

Text-To-Speech Synthesis

Fast Non-Episodic Finite-Horizon RL with K-Step Lookahead Thresholding

2026-01-31 · Jiamin Xu, Kyra Gan arxiv

Online reinforcement learning in non-episodic, finite-horizon MDPs remains underexplored and is challenged by the need to estimate returns to a fixed terminal time. Existing infinite-horizon methods, which often rely on …

Reinforcement Learning

Stable Nonconvex-Nonconcave Training via Linear Interpolation

2023-10-20 · NeurIPS 2023 11 · Thomas Pethick, Wanyun Xie, Volkan Cevher

This paper presents a theoretical analysis of linear interpolation as a principled method for stabilizing (large-scale) neural network training. We argue that instabilities in the optimization process are often caused by…

Causal Attention with Lookahead Keys

2025-09-09 · Zhuoqing Song, Peng Sun, Huizhuo Yuan, Quanquan Gu arxiv

In standard causal attention, each token's query, key, and value (QKV) are static and encode only preceding context. We introduce CAuSal aTtention with Lookahead kEys (CASTLE), an attention mechanism that continually upd…