I tried to get some outputs similar to IBM Confusion Network but still no chance.
It would be great if anyone can suggest Is there any output feature for getting something similar to IBM for Confusion Network/word alternatives in deep speech?
Do you mind explaining a little bit ? The link is not really helpful,
Sure. I need output similar to a weighted directed graph with the peculiarity that each path from the start node to the end node goes through all the other nodes.
I have the same output from the IBM API, but having a similar feature is very useful in deep speech.
It would be great if you could let me know if I can get the same output from deep speech.
You can find more information here
What you are asking for is a lot of non trivial and quite specialized work. I don’t think it’s impossible and it looks interesting, but we definitely don’t have the bandwidth.
@lissyx
Thank you.
If I could have lattice representation output from DeepSpeech I can implement this feature.
I’m sorry, but this is really not something we can do right now, though you are welcome to experiment. It’s likely you can produce that at the CTC decoder level.