MFLA adds finite look-ahead attention plus a CIF-based token counter to Whisper, enabling streaming recognition with a wait-k latency-quality trade-off.
Architecture Experiment We implement the Streaming-Whisper framework in various scale architectures, including Small, Medium, Large-V3 and Large-V3-Turbo
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MFLA: Monotonic Finite Look-ahead Attention for Streaming Speech Recognition
MFLA adds finite look-ahead attention plus a CIF-based token counter to Whisper, enabling streaming recognition with a wait-k latency-quality trade-off.