What's changed: Added per-section figures (cert-figure-retrofit). New AI-901 Chapter 6 filling the missing Domain 2 sub-areas: implementing text and speech (Azure AI Language; Azure AI Speech = STT/TTS/translation; multimodal models for spoken prompts) and implementing computer vision and image generation (multimodal/Azure AI Vision = OCR/object detection; image-generation models). Appended as ch6 to keep existing ch1-5 relatedSections non-breaking.
6.1Implementing Text and Speech
Understand the two routes to implementing text and speech in Foundry—task-focused Foundry Tools (Azure AI Language / Azure AI Speech) and multimodal models that take audio directly—plus text analysis, speech-to-text (STT), text-to-speech (TTS), speech translation, and how to respond to spoken prompts.
Beyond generative apps and agents, Domain 2 expects you to implement text and speech solutions on Foundry. There are two main routes: (1) task-focused services—Azure AI Language (text analysis) and Azure AI Speech (speech) within Foundry Tools (formerly Azure AI services), which you connect to a project and call; and (2) multimodal models—generative models that accept audio/text directly, which you deploy and call. A useful split: "a fixed analysis (sentiment, transcription) → a dedicated service" vs "conversational spoken Q&A → a multimodal model."
6.1.1Implementing text analysis (Azure AI Language)
To process the meaning of text in a structured way, use the task-focused Azure AI Language. It is pretrained and provides the main text-analysis tasks with no model training: sentiment analysis (positive/negative), key phrase extraction, entity recognition (names, places, dates—NER), language detection, summarization, and PII detection. In Foundry you connect this service to a project and call it via endpoint + key (or Microsoft Entra ID auth) to build a lightweight text-analysis app. For fixed tasks it is cheaper, faster, and more predictable than a generative model.
6.1.2Implementing speech (Azure AI Speech)
For audio, use Azure AI Speech (part of Foundry Tools). Its main capabilities are speech recognition (Speech to Text/STT—transcribe spoken audio), speech synthesis (Text to Speech/TTS—read text aloud in a natural voice), speech translation (convert speech into another language), and custom neural voice to craft a tailored voice. For example, transcribing a meeting → STT; auto-generating narration → TTS. In Foundry you connect Speech to a project and build a lightweight speech app.
6.1.3Responding to spoken prompts with a multimodal model
A second route is multimodal models. Deploy a generative model that accepts audio and images directly (not just text), and it can respond straight to a user’s spoken prompt. Where the dedicated Speech service performs fixed conversions (transcribe, read aloud), a multimodal model understands conversationally and responds flexibly. For AI-901, remember: to respond to spoken prompts, choose a deployed multimodal model.
| Goal | How | Capability / note |
|---|---|---|
| Detect sentiment/intent of text | Azure AI Language | Sentiment, NER, summarization (pretrained) |
| Transcribe meeting audio | Azure AI Speech (STT) | Speech to Text = audio → text |
| Read text aloud | Azure AI Speech (TTS) | Text to Speech = text → audio |
| Speak in another language | Azure AI Speech (translation) | Speech translation |
| Converse by voice | Multimodal model | Respond to spoken prompts directly |
About Foundry Tools: at Ignite 2025 the task-focused AI services (Language, Speech, Vision, Content Understanding, etc.) were organized under the name Foundry Tools (formerly Azure AI services). They are pretrained and connected to a Foundry project. This book uses the new name Foundry Tools while still calling each service Azure AI Language / Azure AI Speech as before.
Watch the mix-ups: (1) text analysis (Azure AI Language) vs speech (Azure AI Speech)—split by whether the input is text or audio. (2) Don’t swap STT (audio→text) and TTS (text→audio). (3) Don’t confuse a fixed conversion (dedicated Speech service) with a conversational response (multimodal model). (4) Language/Speech are pretrained, no retraining needed (optional custom features like custom voice exist).
Common: "scenario → how." E.g., "judge sentiment of support text" = Azure AI Language; "transcribe a recording" = Azure AI Speech (STT); "read an article aloud" = Azure AI Speech (TTS); "answer a spoken question directly" = a deployed multimodal model. All are implemented by connecting/deploying within a Foundry project.
6.1.4Section summary
- Text analysis = Azure AI Language (sentiment, key phrase, NER, language detection, summarization, PII; pretrained)
- Speech = Azure AI Speech: STT (audio→text) / TTS (text→audio) / speech translation / custom neural voice
- To respond conversationally to spoken prompts, use a deployed multimodal model
- Task-focused services live under Foundry Tools (formerly Azure AI services), connected to a project
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. Which Azure AI Speech capability transcribes spoken audio into text?
Q2. Which task-focused, pretrained service best analyzes sentiment, key phrases, and entities in text?
Q3. What best implements responding conversationally and directly to a user’s spoken question (audio prompt)?
Q4. Which capability reads article text aloud in a natural voice?
Q5. Which is true of the Language and Speech services in Foundry Tools (formerly Azure AI services)?
Q6. In a call center you must "transcribe recorded audio, then judge the sentiment of its content." Which combination fits best?
Keep track of your progress
The full study guide is free to read. Sign up free to practice with the question bank, track what you have read, review your mistakes, and highlight passages.