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

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Xia, Yinfeng, Li, Huiyan, Le, Chenyang, Wang, Manhong, Sun, Yutao, Ma, Xingyang, Qian, Yanmin
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915324546252800
author Xia, Yinfeng
Li, Huiyan
Le, Chenyang
Wang, Manhong
Sun, Yutao
Ma, Xingyang
Qian, Yanmin
author_facet Xia, Yinfeng
Li, Huiyan
Le, Chenyang
Wang, Manhong
Sun, Yutao
Ma, Xingyang
Qian, Yanmin
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2506_03722
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MFLA: Monotonic Finite Look-ahead Attention for Streaming Speech Recognition
Xia, Yinfeng
Li, Huiyan
Le, Chenyang
Wang, Manhong
Sun, Yutao
Ma, Xingyang
Qian, Yanmin
Computation and Language
Sound
Audio and Speech Processing
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.
title MFLA: Monotonic Finite Look-ahead Attention for Streaming Speech Recognition
topic Computation and Language
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2506.03722