Skip to content
Snippets Groups Projects
Select Git revision
  • master default protected
1 result
You can move around the graph by using the arrow keys.
Created with Raphaël 2.2.04Mar22Feb17529Jan23725Nov231187630Sep242210853add auto break after 10 secondsmastermasternew option for adaptation statedisable on-the-fly adaptationreenable acoustic state saving after endpointdisable lattice generation, change partial hyp intervaladd support for threaded nnet2 decodingFrees memory under LM rescoring related objects when finalizingImplemented optional rescoring with (large) a 'constant ARPA' LMImplemented optional rescoring with (large) a 'constant ARPA' LMAdded decoder.FinalizeDecoding() to make final lattice decoding fasterMade chunk length configurable, can be increased to e.g. 0.2 to get some speedup when multithreaded BLAS is usedMerge pull request #2 from amitbeka/inverse-scaleMerge pull request #3 from amitbeka/gitignoreadd LM scaling propertyadd inverse-scale propertyadd .gitignore with makefile output filesFix for adaptation state being get_property() is called soon after initialization (https://github.com/alumae/gst-kaldi-nnet2-online/issues/1)Added adaptation-state property, needs fairly recent Kaldi to compileAdded adaptation-state property, needs fairly recent Kaldi to compileAdaptation state is saved for the next utterance regardless of the confidenceExperimenting with confidence threshold for updating adaptation stateNot generating lattice if less than 0.1 seconds were processedsaving adaptation state between utterances of a single sessionsmall bugadded python GUI demodoc changesdoc changescosmetic changescosmetic changescosmetic fixesGuard against EOS when not yet decodingcosmetic improvementsnow supports different sample ratescosmetic fixesminor fixminor fixfirstfirstfirst