Overlapping chunks plus a merging rule improve Transformer-based capitalization and punctuation restoration for speech transcripts, but the claimed edge over prior systems is not directly measured.
As the result, the output of ASR con- tains purely a sequence of words or alphabet characters depend- ing on the model type
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Fast and Accurate Capitalization and Punctuation for Automatic Speech Recognition Using Transformer and Chunk Merging
Overlapping chunks plus a merging rule improve Transformer-based capitalization and punctuation restoration for speech transcripts, but the claimed edge over prior systems is not directly measured.