Glossary
Contact-centre AI, in plain words.
The vocabulary of contact centres and speech AI, explained simply and tied to how Intella Studio uses each term.
Business Process Outsourcing
A company that runs customer contact for other companies. Intella Studio is built for BPOs that serve many clients at once.
Automatic quality assurance
Scoring every call against a scorecard automatically, instead of a manual sample of 1–3%.
Scorecard
The list of checks a call is marked against. In a BPO, each client usually has its own.
Calibration
Comparing scores from Auto-QA, your reviewers and your client on the same calls, to keep scoring consistent.
Customer Satisfaction
How satisfied a customer was with a conversation, from a survey or predicted from the conversation itself.
Average Handle Time
The average time a conversation takes, including the work after it ends.
First-Contact Resolution
The share of requests resolved in a single conversation, with no call back.
Service level
The share of calls answered within a target time, usually set in the client’s contract.
Automatic Speech Recognition
Turning spoken audio into text. Intella's ASR understands 25+ Arabic dialects at 95.73% accuracy.
Speech-to-Text
Another name for ASR — the step that transcribes what was said before anything else can happen.
Text-to-Speech
Generating natural spoken audio from text. It's what gives an AI agent a human-like Arabic voice.
Voice-to-voice
A fully spoken loop — speech in, speech out — so a caller talks and is answered in natural speech, no typing.
Word Error Rate
The standard accuracy metric for transcription; lower is better. Intella scores 4.27% WER on Arabic (95.73% accuracy).
Speaker diarization
Working out "who spoke when" in a recording — separating the agent from the customer, turn by turn.
Modern Standard Arabic
Formal, written Arabic used in news and documents — but rarely in everyday speech. Global models over-rely on it.
Dialect & the "dialect gap"
The real spoken Arabic of each region (Egyptian, Gulf, Levantine…). The "dialect gap" is where generic AI breaks down — and where Intella is built to win.
Natural Language Processing
The broad field of machines working with human language — the umbrella over transcription, understanding and generation.
Natural Language Understanding
Extracting meaning from text — intent, entities and sentiment — so the system grasps what a person actually wants.
Intent recognition
Identifying the goal behind an utterance, like "cancel my subscription" or "check my balance."
Entities & slot-filling
Pulling the structured details out of speech — dates, amounts, names — that an action needs to complete.
Large Language Model
A powerful general model for understanding and generating language. Intella fine-tunes LLMs for Arabic and per industry.
Small Language Model
A compact, faster, cheaper model for focused tasks — used to hit the right balance of cost, latency and accuracy.
Fine-tuning
Adapting a base model to a specific domain, industry or brand voice using targeted data.
Retrieval-Augmented Generation
Letting the model answer from your own knowledge base instead of guessing — so responses cite your policies, not the open internet.
Knowledge base
Your documents, policies and FAQs that the AI retrieves from to ground its answers.
Guardrails
Safety and policy limits that keep AI responses compliant, on-brand and within scope — with human review for sensitive cases.
Hallucination
When a model states something false with confidence. RAG and guardrails are how you keep it in check.
Sentiment & emotion analysis
Reading tone and emotion per turn — the trigger for escalation, agent assist and early churn signals.
Intelligent Virtual Agent
An AI agent that reasons and takes real actions — not a scripted bot. Intella Studio AI Agents is an IVA.
Interactive Voice Response
The traditional phone menu ("press 1 for…"). A voice agent replaces it with natural conversation.
Agent assist / co-pilot
Real-time suggestions and compliance nudges shown to a human agent while they're on the call.
Latency
The delay before a response. Intella targets sub-300ms ASR and under one second end-to-end.
Confidence score
How sure the model is about a result — used to decide when to act automatically or route to a human.
Data residency & sovereignty
Keeping data where your rules require and under your control. Intella deploys in the cloud, on-premises or hybrid to meet it.
On-premise deployment
Running the software fully inside your own infrastructure, so sensitive audio never leaves your walls.
See the terms at work
Every term here is a working part of Intella Studio.