1. What This Means in the Exam
Using Azure AI Foundry, you should understand how to:
Accept spoken input (audio)
Process it using AI
Generate a response (text or speech)
Use a multimodal model (handles multiple input/output types)
👉 Think:
“User speaks → AI understands → AI responds”
🧩 2. What is a Multimodal Model?
A multimodal model can work with multiple data types, such as:
Text
Audio (speech)
Images
👉 In this scenario:
Input = speech
Output = text and/or speech
🔄 3. Typical Solution Flow (Exam-Friendly)
User speaks
↓
Speech recognition (audio → text)
↓
Multimodal / LLM processes text
↓
Response generated (text)
↓
(Optional) Speech synthesis (text → audio)
↓
Response returned to user
🗣️ 4. Key Components You Must Recognise
🎤 Speech Recognition
Converts spoken input → text
🤖 Multimodal / Generative Model
Understands input
Generates response
🔊 Speech Synthesis (Optional but common)
Converts response → spoken output
⚙️ Deployment
Model must be deployed in Azure AI Foundry
Accessible via API or application
🎯 5. Example Scenario (Very Likely in Exam)
Scenario:
“A voice assistant allows users to ask questions verbally and hear spoken responses”
Correct understanding:
Input → Speech
Use → Speech recognition
Process → Multimodal/LLM
Output → Speech synthesis
⚖️ 6. How to Identify This in Questions
Clue Meaning
“Spoken input” Speech recognition needed
“Respond verbally” Speech synthesis needed
“Single model handles multiple inputs” Multimodal model
“Voice assistant” Combine all components
⚠️ 7. Common Exam Traps
❌ Thinking multimodal = only images
✔️ Includes speech + text
❌ Forgetting speech recognition step
✔️ Audio must be converted to text
❌ Ignoring speech output
✔️ Often required in scenarios
🧠8. Simple Memory Model
“Hear → Think → Speak”
Hear → Speech recognition
Think → Multimodal model
Speak → Speech synthesis
âś… Summary
To respond to spoken prompts using a deployed multimodal model:
Use Azure AI Foundry
Flow:
Capture speech
Convert to text (recognition)
Process with multimodal model
Generate response
Convert to speech (optional)