AI-102 Question 210
Select 3Your company is building an AI-powered transcription service for customer support calls, which must handle specialized terminology unique to your industry. You need to ensure the speech recognition system can accurately transcribe these terms. Which steps should you take when implementing a custom speech solution in Azure?
- A
Train a custom acoustic model using your industry-specific audio recordings.
- B
Create and upload a custom language model with your industry-specific vocabulary and phrases.
- C
Use the Azure Cognitive Services Speech-to-Text API without any customization.
- D
Test the custom model using real-world audio data and refine it based on performance metrics.
- E
Integrate the custom model into your application using the Custom Speech endpoint.
Show answer and explanation
Correct answers: B, D, E
Explanation
To handle industry-specific terminology, you need to create a custom language model that includes the relevant vocabulary and phrases. Testing the model with real-world audio data ensures its accuracy, and integrating it into your application via the Custom Speech endpoint enables its use in production. Custom acoustic models are not required unless there are unique pronunciation challenges, and generic Speech-to-Text APIs won't capture specialized terms effectively.
- A. Incorrect.
Custom acoustic models are used to handle specific accents or variations in pronunciation, not terminology or vocabulary. For this scenario, a custom language model is more appropriate.
- B. Correct.
Creating and uploading a custom language model is necessary to incorporate your industry-specific vocabulary and phrases for accurate transcription.
- C. Incorrect.
Using the Azure Cognitive Services Speech-to-Text API without customization will not handle specialized terminology accurately, as it relies on general-purpose models.
- D. Correct.
Testing the custom model using real-world data and refining it ensures that the solution meets performance requirements and provides high accuracy.
- E. Correct.
Integrating the custom model into your application using the Custom Speech endpoint is the final step to make the model operational for your transcription service.