0

I'm using Google Speech-to-Text api in order to transcribe phone calls in Hebrew.

Most of the phone calls contains customers that tells their phone-number, can I make some custom class in-order to format these numbers with the correct way?

Other example can be formatting an order-id which has a specific format.

I've read this article https://cloud.google.com/speech-to-text/docs/adaptation-model#custom_classes which tells that it can be a list of items...

What is the difference between class & phrase list?

felixmosh
  • 32,615
  • 9
  • 69
  • 88
  • Ok, you have provided link to custom class description, but what do you mean by "phrase list"? – vitooh Jul 21 '21 at 08:35
  • With model adaption there 2 tools that can help improve predictions, custom classes and phrase list, which looks the same to me... both of them are lists of phrases which can inside the transcribing audio. – felixmosh Jul 21 '21 at 09:03
  • There is nothing about "phrase list" in the documentation. Are you referencing to `PhraseSet` maybe? – vitooh Jul 21 '21 at 12:01
  • Correct, sorry for the confusion – felixmosh Jul 21 '21 at 12:02

1 Answers1

0

PhraseSet indeed contain field phrases which is list of Phrase objects. The object contain two fields: value and boost (reference).

Now, in field value, is the phrase itself and is string. But instead of string value you can define and use CustomClass there, which in fact is a list of phrases (reference). So it works, when you want to add the same boost value to whole list of items. Example from the documentation mentioned by you is one of the best:

For example, you want to transcribe audio data that is likely to include the name of any one of several hundred regional restaurants

Without custom class you would have to add all hundred names with seperate boost value. Instead you can create a list of values using CustomClass and assign one boost value for all of them. Additionally CustomClass can be managed independently to PhreaseSet.

vitooh
  • 4,132
  • 1
  • 5
  • 16
  • Cool, thank you for the explanation, how about the first question? there is any way to create a formatting custom class like google's prepared classes for different languages? – felixmosh Jul 21 '21 at 12:52
  • I am not sure what do you mean " google's prepared classes for different languages"? – vitooh Aug 09 '21 at 06:11
  • these https://cloud.google.com/speech-to-text/docs/class-tokens – felixmosh Aug 09 '21 at 10:35
  • I haven't found anything like this for Herbrew. You can try to request it via [Public Issue Tracker](https://cloud.google.com/speech-to-text/docs/getting-support#file_bugs_or_feature_requests) – vitooh Aug 10 '21 at 12:44