A single RNN-T model with chunked context and consistency regularization improves low-latency streaming ASR accuracy while keeping offline performance intact.
Table 2:Average WER (%) on Open ASR Leaderboard for dif- ferent training configurations of KLD teacher, KLD weightλ, and offlineαfor the same unified RNNT L-size model
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Reducing the Offline-Streaming Gap for Unified ASR Transducer with Consistency Regularization
A single RNN-T model with chunked context and consistency regularization improves low-latency streaming ASR accuracy while keeping offline performance intact.