robotemi/sdk

Temi Automatic Speech Recognition

prashantsmp opened this issue · 2 comments

Is there any other work around to improving default automatic speech recognition?.
Default speech recognition not working properly for the local english slang.

temi ASR is powered by Google speech to text. There is no very good way to do per device adaption as I know.

I only know something can be done is to provide Phrases or PhraseSets to improve the recognition of words like brand names.

It can be set in advance and used the preset classes or sets on the go.
https://cloud.google.com/speech-to-text/docs/adaptation-model#improve_recognition_of_words_and_phrases

It also supports adding phrased to every recognition request.
https://cloud.google.com/speech-to-text/docs/speech-to-text-requests#:~:text=combination%20of%20words.-,speechContext,-%2D%20(optional)%20contains%20additional

This isn't implemented in temi, and I don't know how good it is.

In your case, are you using English as system language or Arabic as system language?

If you use Arabic, maybe language code shall be set as ar-AE to better recognize local English slangs. Currently temi is not using ar-AE when system language is Arabic.

I'm using English as system language.