How to add Phrase List to Speech To Text Golang SDK OR how to create one endpoint for multiple Custom Speech Models?

arvind prajapati 5

I've implemented Azure Cognitive service using Speech To Text Golang SDK. I wanted to add a custom phrase list, but there is no implementation mentioned in this link to add Custom Phrases in Golang. https://learn.microsoft.com/en-us/azure/cognitive-services/speech-service/improve-accuracy-phrase-list?tabs=terminal&pivots=programming-language-javascript

I have created a custom speech model for the "en-us" language, but this endpoint is transcribing the "en-us" language only and not transcribing the speech of other languages.

Is there any option to,

Add custom phrases using Golang SDK.
Use the endpoint of the custom speech model which also transcribes languages other than the "en-us"?

romungi-MSFT 41,961 Reputation points Microsoft Employee

2023-01-11T13:16:46.2066667+00:00

@AKSHAY CHATURVEDI I think GO SDK does not have an implementation of phrase lists. I am trying to check for more information to understand if this is still possible.

For the endpoint scenario, are you planning to use the endpoint of custom speech for language other than the one used for your project while creation? I think you have an option to use the same endpoint to deploy a new model with the same locale or if the base model expires you can re-deploy a new model to the same endpoint. I don't think the endpoint currently supports using more than one project data or locale.
arvind prajapati 5 Reputation points

2023-01-11T14:00:21.39+00:00

Actually, the client wants that Speech SDK to recognize the company name, but as you said, there is no implementation in the SDK of Golang for adding a custom phrase. That's why I created the custom model. So how can I create a new custom model endpoint for a different language which also recognizes the custom word from the old custom model?
romungi-MSFT 41,961 Reputation points Microsoft Employee

2023-01-12T10:08:00.6633333+00:00

I can confirm that the GoLang SDK v1.25.0 will have support for phrase list. Please see this page for announcements. It should be released by the end of this month as per current schedule. For a different language you will have to create a new project and train with the data that supports this language. When you deploy a new endpoint for a new project then the custom model id is used as parameter for calling the endpoint. Using these model ids as parameters in your application you can choose to call the required endpoint based on language detected.
arvind prajapati 5 Reputation points

2023-01-12T13:10:54.72+00:00

Thank you for the information. It'll be really good if it happens. So till then, as per my understanding, I have to create a new endpoint for each language and use them as per the requested language, right? Can I link one endpoint with another so that I won't need to add all the endpoints to the list?
romungi-MSFT 41,961 Reputation points Microsoft Employee

2023-01-13T04:54:48.8466667+00:00

Yes, you cannot link the endpoints. You can however use the base URL and append the custom model ids to call the correct endpoint from your application.
arvind prajapati 5 Reputation points

2023-01-13T13:18:40.1533333+00:00

Can you share an example of it?
romungi-MSFT 41,961 Reputation points Microsoft Employee

2023-01-20T08:57:53.29+00:00

For example, once you deploy the model the endpoint for REST API and SDK would provide the following details for endpoint.

https://westeurope.stt.speech.microsoft.com/speech/recognition/conversation/cognitiveservices/v1?cid=99d47928-887e-4996-xxxx-xxxxxxxxxxxx

The value passed to parameter cid can be different for different models that are deployed for the same region. You can read these values from a variable and set the value as per language used and call the endpoint.

If you are using the SDK this can be directly set using the config.endpoint_id

config = speechsdk.SpeechConfig(subscription="YourSubscriptionKey", region="YourServiceRegion") config.endpoint_id = "YourEndpointId" reco = speechsdk.SpeechRecognizer(speech_config=config)
arvind prajapati 5 Reputation points

2023-01-24T05:46:41.9133333+00:00

@romungi-MSFT Thank you so much for giving the example. I thought your code would be different, but I'm doing the same thing in my code.