Voice AI Glossary: 50 Terms You Need to Know

If you have ever read a voice AI vendor's website and felt lost in a sea of acronyms and technical terms, you are not alone. The voice AI industry loves jargon. STT, TTS, LLM, SIP, PSTN, NLU, AHT, HIPAA... the list goes on.

This glossary exists to help you cut through the confusion. Whether you are evaluating voice AI platforms, sitting in a sales demo, or reading technical documentation, you will find clear, plain-language definitions for the terms that matter most.

We have organized these 50 terms into six categories: Core Concepts, Speech Technology, AI and Machine Learning, Telephony, Business Metrics, and Compliance. Bookmark this page. You will come back to it.

Core Concepts

These foundational terms describe what voice AI is and how it fits into the broader technology landscape.

1. Voice AI

Voice AI refers to artificial intelligence systems that can understand spoken language, process the meaning, and respond with natural-sounding speech in real time. It combines speech recognition, natural language processing, and text-to-speech to enable human-like phone conversations. Voice AI powers virtual agents that can answer calls, book appointments, qualify leads, and handle customer service inquiries without human intervention.

2. Conversational AI

Conversational AI is the broader category of AI technology that enables machines to have natural, human-like conversations. This includes both voice-based systems (voice AI) and text-based systems (chatbots). The key distinction is that conversational AI understands context, remembers previous exchanges, and adapts responses rather than following rigid scripts.

3. IVR (Interactive Voice Response)

IVR is the traditional phone system technology that greets callers with menus: "Press 1 for sales, press 2 for support." IVR systems follow pre-programmed decision trees and cannot understand natural speech beyond basic keyword recognition. Many businesses are replacing IVR systems with voice AI to provide more natural, flexible customer experiences.

4. Virtual Agent

A virtual agent is an AI-powered assistant that handles customer interactions without human involvement. In the voice AI context, virtual agents answer phone calls, understand caller requests, and complete tasks like scheduling appointments or providing information. They differ from chatbots in that they communicate through spoken conversation rather than text.

5. Voice Assistant

A voice assistant is software that responds to voice commands to perform tasks or answer questions. Consumer examples include Siri, Alexa, and Google Assistant. In business contexts, voice assistants are customized AI agents trained on company-specific knowledge to handle customer interactions.

6. Agent Orchestration

Agent orchestration refers to managing multiple AI agents that work together to handle complex tasks. For example, a voice AI system might route a caller to a scheduling agent, then transfer to a billing agent, with seamless handoffs between specialized assistants. Orchestration ensures the right agent handles each part of the conversation.

7. Human Handoff

Human handoff is the process of transferring a caller from an AI agent to a human representative. Good voice AI systems recognize when situations exceed AI capabilities and smoothly transition the conversation to a human agent, providing context so the caller does not need to repeat themselves.

8. Voice Bot

Voice bot is another term for a voice-enabled AI assistant that handles phone conversations. The term is sometimes used interchangeably with "voice AI agent" or "virtual voice agent." Some consider "voice bot" slightly dated terminology, as modern systems are sophisticated enough that "bot" undersells their capabilities.

Speech Technology

These terms describe the technical components that enable machines to hear, understand, and speak.

9. STT (Speech-to-Text)

Speech-to-Text is the technology that converts spoken words into written text. When you speak to a voice AI system, STT is the first step in the pipeline, transcribing your words so the AI can process them. Also called speech recognition or voice recognition. Common STT providers include Deepgram, Google Speech, and Whisper.

10. TTS (Text-to-Speech)

Text-to-Speech is the technology that converts written text into spoken audio. After a voice AI generates a response, TTS converts that text into natural-sounding speech that the caller hears. Modern TTS voices from providers like ElevenLabs and Cartesia sound remarkably human, with natural rhythm, emotion, and intonation.

11. ASR (Automatic Speech Recognition)

Automatic Speech Recognition is the technical term for technology that identifies and transcribes spoken language. ASR and STT are often used interchangeably. ASR systems must handle diverse accents, background noise, and speaking styles to accurately capture what callers say.

12. NLU (Natural Language Understanding)

Natural Language Understanding is the AI technology that interprets the meaning behind spoken or written language. NLU goes beyond recognizing words to understanding intent, context, and nuance. When a caller says "I need to move my appointment," NLU determines that the intent is rescheduling, not physically relocating.

13. Wake Word

A wake word is a specific phrase that activates a voice assistant. Consumer examples include "Hey Siri" or "Alexa." In business voice AI, wake words are less common since systems typically answer phone calls directly rather than waiting for activation phrases.

14. Voice Cloning

Voice cloning uses AI to create a synthetic voice that sounds like a specific person. Businesses sometimes clone a recognizable company voice for brand consistency. The technology requires sample recordings and raises ethical considerations around consent and potential misuse.

15. Voice Biometrics

Voice biometrics uses unique characteristics of a person's voice for identification or authentication. Like a fingerprint, each person's voice has distinctive qualities. Some voice AI systems use voice biometrics to verify caller identity without requiring passwords or security questions.

16. Latency

Latency is the delay between when you speak and when you hear a response. In voice AI, low latency is critical for natural conversations. Delays longer than one to two seconds feel awkward and unnatural. High-quality voice AI platforms achieve sub-second latency to maintain conversational flow.

17. Voice Activity Detection (VAD)

Voice Activity Detection is technology that determines when someone is speaking versus when there is silence or background noise. VAD helps voice AI know when a caller has finished speaking and it is time to respond, preventing awkward interruptions or premature responses.

18. Endpointing

Endpointing refers to detecting when a speaker has finished their turn in conversation. Good endpointing recognizes natural pauses within sentences (where the speaker is not done) versus actual turn endings (where the speaker expects a response). Poor endpointing leads to the AI interrupting or responding prematurely.

AI and Machine Learning Terms

These terms describe the artificial intelligence technologies powering voice AI systems.

19. LLM (Large Language Model)

Large Language Models are AI systems trained on massive amounts of text data to understand and generate human language. LLMs like GPT-4, Claude, and Gemini power the "brain" of voice AI systems, enabling them to understand context, answer questions, and generate natural responses. The quality of the LLM significantly impacts voice AI performance.

20. NLP (Natural Language Processing)

Natural Language Processing is the field of AI concerned with enabling computers to understand, interpret, and generate human language. NLP encompasses technologies like speech recognition, natural language understanding, sentiment analysis, and text generation. It is the umbrella term for making AI language-capable.

21. Fine-tuning

Fine-tuning is the process of taking a pre-trained AI model and training it further on specific data to improve performance for a particular use case. For voice AI, you might fine-tune a language model on your company's documentation, terminology, and customer interaction patterns to improve accuracy and relevance.

22. Prompt Engineering

Prompt engineering is the practice of crafting instructions and context that guide AI behavior. In voice AI, your system prompt tells the AI assistant who it is, what it knows, how it should behave, and what tasks it should perform. Effective prompt engineering is crucial for voice AI performance.

23. RAG (Retrieval Augmented Generation)

Retrieval Augmented Generation is a technique where AI retrieves relevant information from a knowledge base before generating a response. Instead of relying solely on its training data, the AI searches your documents, FAQs, or databases to find accurate, up-to-date information. RAG enables voice AI to answer questions about your specific business.

24. Hallucination

In AI, hallucination refers to when a model generates information that sounds plausible but is factually incorrect or fabricated. Voice AI systems can hallucinate if asked questions beyond their knowledge. Techniques like RAG and careful prompt engineering help reduce hallucinations by grounding responses in verified information.

25. Intent Recognition

Intent recognition is identifying what a user is trying to accomplish from their spoken or written input. When a caller says "What time do you close?" the intent is seeking business hours information. Voice AI systems classify intents to route conversations and determine appropriate responses.

26. Entity Extraction

Entity extraction identifies and extracts specific pieces of information from natural language. Entities might include names, dates, times, phone numbers, addresses, or product names. When a caller says "I need an appointment next Tuesday at 2pm," entity extraction captures "next Tuesday" and "2pm" as scheduled time entities.

27. Sentiment Analysis

Sentiment analysis determines the emotional tone of spoken or written language, identifying whether someone is happy, frustrated, angry, or neutral. Voice AI can use sentiment analysis to detect upset callers and adjust its approach or escalate to human agents when emotions run high.

28. Context Window

Context window refers to the amount of previous conversation an AI can consider when generating responses. A larger context window means the AI can remember more of the conversation history, leading to more coherent and contextually appropriate responses. LLMs have varying context window sizes.

29. Token

A token is the basic unit of text that AI models process. Tokens might be words, parts of words, or punctuation. AI pricing often involves cost per token, and context windows are measured in tokens. Understanding tokens helps you estimate AI costs and capabilities.

Telephony Terms

These terms describe the technology that connects voice AI to phone networks and enables calls.

30. SIP (Session Initiation Protocol)

SIP is the standard protocol for initiating, maintaining, and terminating voice and video calls over the internet. SIP trunks connect voice AI platforms to phone networks, enabling them to make and receive calls. If you bring your own telephony to a voice AI platform, you are likely using SIP.

31. SIP Trunk

A SIP trunk is a virtual phone line that connects your phone system to the public phone network over the internet. Businesses use SIP trunks for cost-effective calling without traditional phone lines. Voice AI platforms can connect to your existing SIP trunks or provide their own.

32. PSTN (Public Switched Telephone Network)

The Public Switched Telephone Network is the traditional phone infrastructure connecting landlines and mobile phones globally. When someone calls a voice AI agent from their regular phone, that call travels through the PSTN before reaching the AI system. PSTN access requires telephony providers.

33. VoIP (Voice over Internet Protocol)

VoIP is technology that transmits voice calls over the internet rather than traditional phone lines. Services like Zoom, Microsoft Teams, and Google Voice use VoIP. Voice AI platforms typically use VoIP for handling calls, converting between VoIP and PSTN as needed.

34. Twilio

Twilio is a cloud communications platform that provides APIs for voice calls, SMS, and other communication channels. Many voice AI platforms integrate with or compete with Twilio for telephony services. Twilio is often mentioned in voice AI contexts as a telephony provider option.

35. WebRTC

WebRTC (Web Real-Time Communication) is technology enabling audio and video communication directly through web browsers without plugins. Some voice AI applications use WebRTC for web-based voice interactions, enabling customers to speak with AI agents through website widgets.

36. Call Recording

Call recording captures audio of phone conversations for quality assurance, training, compliance, or documentation purposes. Voice AI platforms typically offer automatic call recording along with transcripts. Regulatory requirements often mandate disclosure and consent for call recording.

37. Call Transfer

Call transfer moves an active call from one destination to another. Voice AI systems use warm transfers (where context is provided before handoff) and cold transfers (direct transfer without context). Smooth call transfers are essential when AI agents need to escalate to human representatives.

38. DID (Direct Inward Dialing)

DID is a telephony feature that assigns individual phone numbers to specific destinations without requiring separate physical phone lines. Voice AI platforms use DID to provide dedicated phone numbers that route directly to AI agents.

39. Caller ID

Caller ID identifies the phone number and sometimes the name of an incoming caller. Voice AI systems can use Caller ID to personalize greetings, look up customer information, and route calls appropriately. For outbound calls, configuring the displayed Caller ID affects answer rates.

Business Metrics

These terms describe how businesses measure voice AI performance and customer service effectiveness.

40. CSAT (Customer Satisfaction Score)

Customer Satisfaction Score measures how satisfied customers are with their experience, typically through post-interaction surveys asking customers to rate satisfaction on a scale. Voice AI success is often measured by comparing CSAT scores between AI-handled and human-handled calls.

41. AHT (Average Handle Time)

Average Handle Time measures the total duration of customer interactions, including conversation time and any after-call work. Voice AI typically reduces AHT by handling routine inquiries quickly. However, optimizing purely for speed can negatively impact customer experience.

42. FCR (First Call Resolution)

First Call Resolution measures the percentage of customer issues resolved in a single interaction without requiring callbacks or transfers. High FCR indicates effective service. Voice AI can improve FCR by providing instant access to information and completing transactions in one call.

43. Containment Rate

Containment rate measures the percentage of calls fully handled by voice AI without requiring human intervention. A higher containment rate indicates the AI successfully resolves more customer needs independently. However, some calls appropriately require human assistance.

44. Deflection

Deflection refers to redirecting potential phone calls to self-service options like voice AI, chatbots, or online resources. Deflection reduces call center volume and costs. However, excessive or poorly implemented deflection frustrates customers who genuinely need assistance.

45. Abandonment Rate

Abandonment rate measures the percentage of callers who hang up before their call is answered or their issue is resolved. Traditional call centers see high abandonment during peak times. Voice AI with instant answer capabilities dramatically reduces abandonment by eliminating hold times.

46. ROI (Return on Investment)

Return on Investment measures the financial return generated relative to the cost of an investment. Voice AI ROI calculations typically compare AI costs to the equivalent human staffing costs, plus factor in benefits like 24/7 availability, improved customer satisfaction, and reduced abandonment.

Compliance and Security Terms

These terms describe regulations and standards relevant to voice AI implementations.

47. HIPAA (Health Insurance Portability and Accountability Act)

HIPAA is US healthcare legislation that establishes standards for protecting sensitive patient health information. Voice AI systems handling patient calls must comply with HIPAA requirements including data encryption, access controls, audit trails, and Business Associate Agreements. Healthcare organizations should verify HIPAA compliance before implementing voice AI.

GDPR is European Union regulation governing data privacy and protection for EU residents. Voice AI systems serving European customers must comply with GDPR requirements around consent, data access rights, data portability, and the right to be forgotten. GDPR applies regardless of where the business is located.

49. TCPA (Telephone Consumer Protection Act)

TCPA is US legislation regulating telemarketing calls, auto-dialed calls, pre-recorded calls, and text messages. Voice AI systems making outbound calls must comply with TCPA requirements including consent, calling time restrictions, and do-not-call list compliance. TCPA violations carry significant penalties.

50. SOC 2 (Service Organization Control 2)

SOC 2 is a compliance framework demonstrating that a service provider securely manages data to protect customer privacy. SOC 2 certification requires audits of security, availability, processing integrity, confidentiality, and privacy controls. Many enterprises require SOC 2 compliance from voice AI vendors before implementation.

How to Use This Glossary

Now that you understand these terms, you are equipped to navigate voice AI conversations with confidence. Here are some practical ways to use this knowledge:

During vendor evaluations: When platforms mention STT latency or LLM fine-tuning, you understand what they mean and can ask informed follow-up questions.

Reading documentation: Technical docs become accessible when you know the terminology. RAG implementations, SIP trunk configurations, and intent recognition settings make sense in context.

Internal discussions: Share this glossary with your team so everyone speaks the same language when evaluating and implementing voice AI.

Negotiating contracts: Understanding terms like HIPAA BAA, SOC 2 compliance, and containment rates helps you negotiate appropriate service level agreements.

Still Have Questions?

Voice AI technology evolves rapidly, and new terminology emerges regularly. If you encounter terms not covered here, or need clarification on how specific concepts apply to your situation, reach out.

At Burki, we believe buyers should understand the technology they are purchasing. Our team is happy to explain any technical concepts in plain language and help you determine if voice AI is right for your business.

[Start your free trial at burki.dev/signup](https://burki.dev/signup) to see voice AI in action. You get 200 free minutes and a phone number for 30 days. No credit card required, no jargon-filled sales pitch. Just the technology, ready to test.

Questions about specific voice AI terminology or how these concepts apply to your business? Email [[email protected]](mailto:[email protected]) or explore our [documentation](https://docs.burki.dev) for detailed technical guides.