In the old online setup, we have OnlineFasterDecoder which outputs words/chunks instead of waiting for the entire utterance. Do we have any such decoder in the new framework, where we can have streaming captions? The webpage says that OnlineFasterDecoder might be moved and made compatible with the online2 setup. I was wondering if we had any updates on that?
Thanks
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Class SingleUtteranceNnet2Decoder has a function GetBestPath which is
internally quite lightweight and which can be used at any point to get
the most likely traceback at any point. The only thing it doesn't do
which OnlineFasterDecoder does, is to keep track of when each word
becomes inevitable in the traceback, so that you can produce output
without ever having to backtrack on what you already produced. But
for some purposes this won't matter- you can just display the current
best guess.
Dan
In the old online setup, we have OnlineFasterDecoder which outputs
words/chunks instead of waiting for the entire utterance. Do we have any
such decoder in the new framework, where we can have streaming captions? The
webpage says that OnlineFasterDecoder might be moved and made compatible
with the online2 setup. I was wondering if we had any updates on that?
Thanks
OnlineFasterDecoder like decoder in online2 setup?
In the old online setup, we have OnlineFasterDecoder which outputs words/chunks instead of waiting for the entire utterance. Do we have any such decoder in the new framework, where we can have streaming captions? The webpage says that OnlineFasterDecoder might be moved and made compatible with the online2 setup. I was wondering if we had any updates on that?
Thanks
Class SingleUtteranceNnet2Decoder has a function GetBestPath which is
internally quite lightweight and which can be used at any point to get
the most likely traceback at any point. The only thing it doesn't do
which OnlineFasterDecoder does, is to keep track of when each word
becomes inevitable in the traceback, so that you can produce output
without ever having to backtrack on what you already produced. But
for some purposes this won't matter- you can just display the current
best guess.
Dan
On Sat, Jan 3, 2015 at 4:14 PM, Yash y91@users.sf.net wrote: