kate-egorova Goto Github PK

asr-hybrid-decoding

This is an extension of kaldi speech recognition software which allows to perform decoding of speech with hybrid word and phoneme graphs. The output is a mix of in-vocabulary words and phoneme sequences. This decoding is suitable for systems with only a small dictionary available and for further recovery of OOV words.

beer

oov-recovery-in-hybrid-asr-system

Out-of-vocabulary word recovery system makes use of a hybrid decoding network with both words and sub-word units. In the decoded lattices, candidates for OOV regions are identified as sub-graphs of sub-word units. To facilitate OOV word recovery, we search for recurring OOVs by clustering the detected candidate OOVs.