Post Processing ASR Output

Hi,
Would like to know is there any method or technique to post-process the stt output?
Thanks

Get JSON output and do with it whatever you want :slight_smile:

Hi,

it depends on your use case.

I my project I convert written numbers to digits. Also, since I use very simple alphabet, i change some letters into the specific characters(umlauts) that uses german language.

1 Like

@buxbaum_y what lib do you use for that or is it hand written?

for digits: I tried duckling and lingua_franca (from mycroft). Both are making mistakes in some cases, I’m working currently on it, I would rather choose duckling and implement some workaround for numbers that are problematic.
for umlauts: I tried with regex but it is not a perfect solution. Right now I use hand written function based on the regex and dictionaries.

Do you have some solutions?

1 Like

Same, have tried duckling and going with some specialized scripts for now. Thanks for sharing.

1 Like

@othiele sir, is it possible to have a look on your script for converting digits (post processing)? I’m would be very thankful because to be honest I’m struggling a lot with this feature…

Unfortunately no, we did this for a larger client who doesn’t want the scripts out in the open.

1 Like

I have a model for a specific client and the same words are not predict correctly.
I also include all those words in lm, but then also there is not much improvement.
Want to know is there a way, through which I can post-process the words, and correct them?

@tanmayjain can you give some examples?