Send expected words in request without model RRS feed

  • Question

  • The thing I like about the Google Speech to Text API, is that it has high accuracy, a fair bit better than the Azure API in my experience. The reason being, I am able to send a list of expected words in my request to the Google API, and therefore, increase the likelihood of those terms/phrases occuring. With Azure, I am restricted to making speech models. The problem with this is, I would have to create a new model for each request, as each request will have a different vocabulary etc. from the rest. The requests are not industry specific, they are completely random and may contain very technical terms from a range of industries/areas.

    For example, using the speech recognition test on the azure website:

    When I say "Farady" (a term used in chemistry), I get "Saturday"

    And when I say "Anaclisis" (some random word I found), I get "Anna classes"

    So, my question is, can I send a list of expected words with my request to the Azure API? Or, how can I use custom models to accomodate for a wide range of industries...?


    Tuesday, May 22, 2018 6:22 AM