Hello, you have done a very good job with the DeepSpeech project.
What I would like to understand is the proceess to add more words to the scorer.
I have created my model and trained it, but later I need to include more words to the scorer that I use.
Do I have to append the new words to the previous file that I have used and run the generate_lm.py and generate_scorer_package again and then upload the new scorer to the server?
Yes that is pretty much what you have to do.
To include new vocabulary you need to include it in your language model binary and in the vocabulary file and than generate a new scorer file from that.
Have you had a look at the Playbook section? https://mozilla.github.io/deepspeech-playbook/SCORER.html
yes, I have followed the Playbook and managed to create the scorer.
But while the previous scorer works, the new one doesn’t recognize not even one word.
It only presents random letters without meaning…
I have made all possible changes, but I cannot figure why the new scorers do now work.
Are you using the same alphabet?
The alphabet was the same, but the bytewise was different from the one used to create the training.
Once the same alphabet was used the machine started to recognize the voice entries.