AI-901
Deep Dive
Azure has four main services for NLP: Language (text analysis), Speech (voice), Translator (translation), and Bot Service (conversation). The AI-901 tests which one to use in each scenario.
Contents
💬
Azure AI Language
Analysis and understanding of written text: sentiment, NER, key phrases, classification, summarization, question answering.
🎤
Azure AI Speech
Voice processing: transcription (STT), synthesis (TTS), speech translation, speaker recognition.
🌍
Azure Translator
Text translation between 100+ languages. Automatic source language detection. Simple REST API.
🤖
Azure Bot Service
Platform to build, host, and connect conversational bots to multiple channels: Teams, web, WhatsApp, etc.
A unified service that groups all text analysis capabilities. They used to be separate services (Text Analytics, LUIS, QnA Maker) — now they're consolidated.
Sentiment Analysis
Detects positive/negative/neutral opinion. Supports opinion mining by aspect.
Key Phrase Extraction
Extracts the main topics and concepts from a text.
Named Entity Recognition
Identifies entities: people, places, dates, organizations, quantities.
Entity Linking
Links entities to their Wikipedia/Bing article for disambiguation.
Language Detection
Identifies the language of a text with a confidence score.
Text Summarization
Extractive (extracts key sentences) or abstractive (generates a new summary).
PII Detection
Detects and optionally redacts personally identifiable information.
Custom Text Classification
Trains a classifier with your own categories using few examples.
Custom NER
Trains a custom entity recognizer for your specific domain.
Prebuilt vs Custom
Prebuilt (no training)
Sentiment, NER, Key Phrases, Language Detection, Entity Linking, PII Detection, Summarization. Works immediately without training data.
Custom (with training)
Custom Text Classification, Custom NER, Conversational Language Understanding, Custom Question Answering. Require labeled training data.
Detect whether a text is positive, negative, or mixed, with a per-sentence breakdown.
+1 more...
Automatically identify people, places, organizations, dates, and quantities in a text.
+1 more...
Identify what language each message is written in across a multilingual support inbox.
+1 more...
CLU (successor of LUIS) lets you train models that understand the intent behind a message and extract specific entities from the conversation.
Intents
What the user wants to do. Each utterance is classified into an intent.
"Book a flight to Madrid"
→ Intent: BookFlight
Entities
The specific data within the message that the system needs to act.
"Book a flight to Madrid"
→ Destination: "Madrid"
Utterances
Examples of how users express each intent. Used to train the model.
"I want to go to Barcelona"
"I need a ticket to Rome"
"Trip to Paris please"
Exam trap: CLU vs Sentiment
CLU understands intents to build conversational bots (what does the user want to do?). Sentiment Analysis detects emotional tone (are they happy or upset?). An order-taking bot uses CLU; a review analysis uses Sentiment.
Lets you create a knowledge base of question-answer pairs, and then the model answers questions in natural language by finding the most suitable answer. Successor of QnA Maker.
CLU vs Question Answering
CLU: the user gives an instruction ("book a flight") → the bot understands the intent and acts.
Question Answering: the user asks a question ("how much does shipping cost?") → the bot searches for and returns the answer from the knowledge base.
Speech-to-Text (STT)
Speech-to-text transcription in real time or in batch. 100+ languages.
Transcribing support calls, meetings, medical dictation.
Text-to-Speech (TTS)
400+ neural voices in 140+ languages. Natural and expressive.
Voice assistants, accessibility, automated announcements.
Speech Translation
STT + Translation + TTS in a single call. Real time.
Multilingual conference calls, live captioning.
Speaker Recognition
Identifies who spoke based on unique vocal characteristics.
Voice authentication, diarization (who spoke when).
Custom Speech
Adapts STT with pronunciations and vocabulary specific to your domain.
Medical terms, proper names, technical jargon.
Custom Neural Voice
Creates a unique synthetic voice for your brand.
Virtual assistants with a consistent brand identity.
Speech Translation vs Azure Translator
Speech Translation (part of Azure Speech): voice → translated text → voice. A single service, real time.
Azure Translator: text → translated text. Text only, no voice.
Convert a support call recording into analyzable text, with a confidence score.
+1 more...
Main capabilities
Document Translation
Translates complete documents (PDF, Word, PowerPoint) while preserving the original format. Ideal for large-scale content localization.
Custom Translator
Uses your own past translations to train a customized model. Useful for industries with specific terminology: legal, medical, technical.
Exam: when to use Translator
Text → Text in another language. If voice is involved → Azure Speech (Speech Translation). If you need to analyze the text → Azure Language.
Translate a message from Spanish to English in under 200 milliseconds.
+1 more...
A platform to build, host, and connect conversational bots. Azure Bot Service does not include understanding AI — it provides the infrastructure and channels. The intelligence comes from CLU, Question Answering, or other services that get integrated.
What Azure Bot Service does
Typical architecture
| Scenario | Service | Specific feature |
|---|---|---|
| Analyze product reviews to know if they are positive or negative | Azure AI Language | Sentiment Analysis |
| Extract doctor and hospital names from clinical reports | Azure AI Language | Custom NER or prebuilt NER |
| Build a bot that understands "I want to order a large pizza" | Azure AI Language | CLU (Conversational Language Understanding) |
| Create an FAQ chatbot for the HR department | Azure AI Language | Question Answering |
| Transcribe call center calls to text | Azure AI Speech | Speech-to-Text (STT) |
| Generate audio for screen readers | Azure AI Speech | Text-to-Speech (TTS) |
| Translate a live conference transcript into 5 languages | Azure AI Speech | Speech Translation |
| Translate 10,000 Word documents into Portuguese | Azure Translator | Document Translation |
| Detect what language a received email is in | Azure AI Language or Translator | Language Detection |
| Deploy a bot on Teams, web, and WhatsApp simultaneously | Azure Bot Service | Multi-channel deployment |
❓ Are LUIS and QnA Maker still separate services?
No. LUIS was replaced by CLU (Conversational Language Understanding) and QnA Maker was replaced by Question Answering — both are now integrated into Azure AI Language Service. The exam may use the new names or the old ones.
❓ Does Azure Bot Service have built-in AI?
Not directly. Bot Service is the platform/infrastructure for the bot. Language understanding comes from CLU or Question Answering, which are integrated as separate services. Bot Service manages channels, authentication, and hosting.
❓ Can I use Language Detection to know if a text is spam?
No. Language Detection only detects the LANGUAGE of the text. To classify text as spam/not spam you use Custom Text Classification (Azure AI Language). They are completely different capabilities.
❓ Can Azure Speech analyze sentiment from voice?
Azure Speech transcribes voice to text. To analyze sentiment, you would need to pass the resulting text to Azure AI Language → Sentiment Analysis. There is no direct sentiment analysis on audio among the services covered in AI-901.
Did you get this topic?
Test what you just learned
A retail company wants to implement a chatbot on its website and on WhatsApp that: (1) understands when a customer says 'I want to return my order' or 'I need to make a return', and (2) answers questions about the return policy. Which Azure services do they need?
Put what you learned to the test with original AI-901 exam questions.
Start AI-901 mock examUnlimited mock exams with the Pro plan
Log in to track your progress.