A bandwidth embedding, a small learned vector marking speech as narrowband or wideband, improves narrowband word error rate by 13% relative in a single mixed-bandwidth acoustic model without degrading wideband speech.
Evaluations are performed on 54 and 4 hours of wideband and narrowband test sets
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
eess.AS 1years
2019 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Bandwidth Embeddings for Mixed-bandwidth Speech Recognition
A bandwidth embedding, a small learned vector marking speech as narrowband or wideband, improves narrowband word error rate by 13% relative in a single mixed-bandwidth acoustic model without degrading wideband speech.