How can we emit word confidence while decoding?

[aayushkubb]

I am trying to estimate word confidence while doing CTC decoding, one of the simplest ways I could think of is just multiplying the associated character probabilities at that particular time step to get an approximation.

However, a lot of people have been using the concept of posterior probabilities and lattice to get the word confidence.

Is there a simplified way to have word confidence estimation during decoding?