AI-102 Question 248
Select 3You are developing a transcription application using the Azure AI Speech service to convert live conference audio into text. The application must support multiple speakers and accurately attribute transcriptions to each speaker. Additionally, you need to ensure that the service adapts to domain-specific terminology used during the conference. What steps should you take to configure the Speech service for this requirement?
- A
Enable speaker diarization to identify and distinguish between multiple speakers.
- B
Train a custom acoustic model to improve transcription accuracy for domain-specific terminology.
- C
Use a custom language model to include domain-specific vocabulary and phrases.
- D
Enable profanity masking to ensure inappropriate words are excluded from the transcription.
- E
Configure real-time audio streaming to send the conference audio to the Speech service.
Show answer and explanation
Correct answers: A, C, E
Explanation
To meet the requirements of transcribing live conference audio with accurate speaker attribution and domain-specific terminology recognition, you need to enable speaker diarization for identifying multiple speakers, use a custom language model to handle domain-specific vocabulary, and configure real-time audio streaming to send the audio to the Speech service. These steps ensure the service is properly configured for multi-speaker transcription in a specialized context.
- A. Correct.
Enabling speaker diarization allows the Azure AI Speech service to identify and distinguish between multiple speakers, which is essential for accurate speaker attribution in a multi-speaker setting.
- B. Incorrect.
Training a custom acoustic model is typically unnecessary for this scenario unless there are unique acoustic conditions. Instead, using a custom language model is more relevant for domain-specific terminology.
- C. Correct.
Using a custom language model ensures that the Speech service understands and correctly transcribes domain-specific vocabulary and phrases, which is required for your use case.
- D. Incorrect.
While profanity masking is useful in some contexts, it is not directly related to solving the requirements of speaker attribution or domain-specific terminology in this scenario.
- E. Correct.
Real-time audio streaming is necessary to send the live conference audio to the Speech service for transcription, making it a key step in the configuration.