Closed cogmeta closed 5 years ago
works for me :o)
@gooofy, yep i got it working after copying some files. Was it supposed to work without any fixes? I will try again from scratch.
Btw, thank you so much for the amazing work. I just have one final quick question: with a good language model what kind of accuracy one can expect? Any experimentation results on call center/phone record data? Thanks.
glad to hear you got it working :) - it is difficult for me to test whether this scripting setup works out of the box as I am usually using it on the very machine I am developing it on so naturally all required components are there usually. If you have pull request, I am happy to review them.
accuracy by lm adaptation: totally depends on how narrow your domain is - if you can build a narrow language model that matches your domain you will likely see very low error rates while at the same time your model will have less and less resistance to any kind of noise (e.g. out of domain utterances, background noise, etc.).
I have been trying to run the adaptation script. I followed the instructions but the script fails at multiple stages. I tried to correct but then gave up after dozens of errors. Has anybody ran it successfully?