The AI voice agent space is full of acronyms and jargon. This glossary defines the 24 terms you'll encounter when researching, deploying, or managing AI voice agents.
AI Voice Agent
Software that answers or makes phone calls and holds spoken conversations with callers using natural language understanding, speech-to-text, large language models, and text-to-speech technology.
ASR (Automatic Speech Recognition)
The technology that converts spoken audio into text. Also called speech-to-text (STT). Modern ASR achieves 95-98% accuracy on clear speech, lower on accented or noisy audio.
BAA (Business Associate Agreement)
A legal contract required under HIPAA between a healthcare provider and any vendor that processes Protected Health Information (PHI) on the provider's behalf. Required for any AI voice platform used in healthcare.
Concurrency
The number of simultaneous calls an AI voice agent can handle. All major platforms support high concurrency (100+ simultaneous calls), limited mainly by your Twilio account capacity.
DTMF (Dual-Tone Multi-Frequency)
The tones generated by phone keypad presses. Used for PCI-compliant payment capture (caller enters card number via keypad, AI never 'hears' the number) and interactive menus.
FCC (Federal Communications Commission)
The US regulatory body that enforces TCPA and issues rules on AI voice calls, including the 2024 ruling requiring AI disclosure.
Function Calling
A capability where the AI can execute custom functions during a call (check inventory, schedule appointment, look up customer record) via API calls. Supported by all major platforms.
HIPAA (Health Insurance Portability and Accountability Act)
US law protecting Protected Health Information (PHI). Applies to AI voice agents in healthcare settings, requiring BAA with platform and minimum-necessary data collection.
IVR (Interactive Voice Response)
Traditional phone menu systems ('press 1 for sales, press 2 for support'). AI voice agents are the modern replacement for IVR, offering natural language instead of fixed menus.
Latency
The time between caller finishing a sentence and AI starting its response. Modern AI voice platforms achieve 400-700ms median latency, fast enough for natural conversation.
LLM (Large Language Model)
The AI 'brain' that decides what the agent should say next based on the conversation. Examples: GPT-4, Claude, Gemini. Some platforms (Vapi) let you bring your own LLM.
NLP (Natural Language Processing)
The field of AI focused on understanding and generating human language. AI voice agents use NLP for understanding callers and generating responses.
PCI DSS (Payment Card Industry Data Security Standard)
Security standard for handling payment card data. AI voice agents collecting payment must use DTMF (keypad) input rather than voice capture for card numbers.
PHI (Protected Health Information)
Individually identifiable health information protected under HIPAA. Includes patient name + appointment type, symptoms, diagnoses, medication information, and more.
STT (Speech-to-Text)
See ASR. The technology that converts spoken audio into text for the LLM to process.
System Prompt
The instructions given to the AI defining its persona, goals, decision trees, and constraints. The system prompt is the most important configuration for AI voice agents.
TCPA (Telephone Consumer Protection Act)
1991 US law restricting telemarketing calls, auto-dialed calls, prerecorded voice calls, and (since 2024) AI-generated voice calls. Violations cost $500-$1,500 per call.
Text-to-Speech (TTS)
The technology that converts text into spoken audio. Modern TTS (ElevenLabs, Cartesia) produces natural-sounding speech indistinguishable from humans.
Twilio
A cloud communications platform that provides phone numbers and call connectivity for most AI voice platforms. Vapi, Retell, Synthflow, and Bland all use Twilio under the hood.
Voice Cloning
Creating a custom TTS voice from a sample of recorded audio. All major AI voice platforms support voice cloning via ElevenLabs integration. Requires 30+ seconds of sample audio.
Voicemail Detection
The AI's ability to recognize when a call goes to voicemail and either leave a natural-sounding message or hang up. Saves money on outbound campaigns (no point in talking to a voicemail).
Warm Transfer
Transferring a caller to a human while the AI stays on the line briefly to provide context ('I have John Smith on the line calling about a water heater emergency at 123 Main St'). Better experience than cold transfers.
Webhook
An HTTP callback that triggers when an event occurs. AI voice platforms send webhooks for call events (call-started, call-ended, lead-captured), enabling CRM integration and workflow automation.
Zapier
A no-code automation platform that bridges different SaaS apps. Used to integrate AI voice platforms with CRMs that lack native integrations. Costs $29-$100/month for typical usage.