search

jtrmal / kaldi2020

Apache License 2.0
27 stars 3 forks source link

Questions from audience - deployment of ML models #13

Open leibny opened 3 years ago

leibny commented 3 years ago

Would there be a focus not only on the deep learning frameworks but also the deployment of ML models ahead of Kaldi?

kkm000 commented 3 years ago

If by "deployment" you mean building a production system based on the trained model, then very emphatic yes! See also https://github.com/jtrmal/kaldi2020/issues/32#issuecomment-702132424

Unlike many deployment scenarios (e. g., data analytics pipeline), ASR applications are much more diverse. Do you want real-time subtitling of the news, or a customer support robot helping someone with their order, or an in-car command and control system (play this song, navigate there)? These are all "deployments." As I understand, there are no plans for a generic one-fits-all "deployment," chiefly due to the sheer variety of scenarios in which Kaldi is currently used.