AI voice calling glossary

24terms you'll run into while evaluating AI calling software, explained in plain English — no vendor jargon, no assumed background.

AI voice agent: Software that makes or receives phone calls and holds a spoken conversation with the person on the other end. It listens, understands what was said, decides on a reply from its instructions and knowledge, and speaks that reply in a synthetic voice — all in under a second so the call feels natural.
Voice-to-voice latency: The delay between a caller finishing speaking and the agent starting to speak. It is the single most important quality measure in voice AI: humans take roughly 200ms to respond in conversation, so anything above about a second feels broken. Leadvox agents typically respond in 400ms to 800ms.
STT (speech-to-text): The component that converts the caller's audio into text so the AI can understand it. Also called ASR (automatic speech recognition). Accuracy on Indian languages, accents and code-mixed Hindi-English speech is what separates usable STT from unusable STT in this market.
TTS (text-to-speech): The component that converts the AI's text reply into spoken audio. Modern neural TTS produces speech with natural intonation and pacing rather than the flat, clipped delivery of older systems.
LLM (large language model): The model that decides what the agent says next, based on the conversation so far, the persona it was given, and any knowledge base documents attached to it. It is the reasoning layer between speech-to-text and text-to-speech.
Turn detection: Deciding when the caller has actually finished speaking rather than just pausing mid-sentence. Get it wrong in one direction and the agent interrupts; get it wrong in the other and it leaves an awkward silence. Also called endpointing.
Endpointing delay: How long the system waits after the caller stops making sound before treating the turn as finished. A short delay makes the agent fast but prone to interrupting; a long delay makes it patient but slow. Good systems vary this based on whether the sentence sounded complete.
Barge-in: The caller interrupting the agent mid-sentence, and the agent stopping to listen rather than talking over them. Without barge-in support, a call becomes a frustrating monologue.
Warm transfer: Handing a live call to a human while the caller stays on the line, so there is no callback and no repetition. The opposite is a cold transfer, where the caller is told someone will ring them back — which is where most leads are lost.
SIP trunk: The connection that carries phone calls between a software platform and the public telephone network over the internet. It is what allows an AI agent to dial a normal mobile number and be dialled from one.
DID (direct inward dialling) number: A normal phone number allocated to a business that routes incoming calls to a specific destination — in this context, to an AI agent rather than a desk phone.
Outbound calling: Calls the business makes to customers: new lead follow-up, renewal reminders, confirmation calls, win-back campaigns. Most AI calling value in India sits here, because outbound is the volume that human teams can never fully cover.
Inbound calling: Calls customers make to the business. An AI agent answering inbound calls gives consistent answers to routine questions and frees the front desk, transferring anything unusual to a person.
Lead response time: How long between a customer submitting an enquiry and someone actually calling them. It is the strongest single predictor of whether a comparison-shopped lead converts, and the reason instant auto-calling exists.
Lead ads: Ad formats on Facebook and Instagram where the user submits their contact details inside the app without visiting a website. They generate high volume cheaply, which makes fast, automated follow-up essential.
Lead qualification: Establishing whether a lead is actually worth a salesperson's time — budget, requirement, timeline, eligibility. Doing this on an automated first call means expensive human time is spent only on leads that pass.
Hot lead: A prospect whose conversation showed genuine, near-term buying intent. Leadvox labels each completed call so hot leads can be actioned first rather than found by listening to recordings.
Knowledge base: The documents you attach to an agent — brochures, price lists, policy documents, FAQ sheets — that it answers from during a call. It is what stops the agent inventing an answer, and updating one document updates every future call at once.
Agent persona: The plain-text instructions describing who the agent is, how it should speak, what it must ask, and what it must never do — such as never quoting a price or never giving medical advice.
Call transcript: The written record of everything said on a call by both parties. Far faster to review than audio, and what makes it practical to audit hundreds of calls rather than spot-checking three.
WhatsApp message template: A pre-approved message format that a business can send to someone who has not messaged it first. WhatsApp only reliably delivers business-initiated messages to new contacts as templates, which is why automated follow-up is built around them rather than free text.
RTO (return to origin): An e-commerce order that is shipped, refused or undelivered, and returned to the seller. The seller pays shipping both ways and earns nothing, which is why confirming cash-on-delivery orders by phone before dispatch matters so much in India.
COD confirmation call: A short call to a customer who placed a cash-on-delivery order, confirming that they placed it, that the address is right, and that they still want it — made before the parcel is dispatched, so a doubtful order can be cancelled instead of shipped twice.
GEO (generative engine optimisation): Making a website findable and quotable by AI answer engines such as ChatGPT, Claude, Perplexity and Google AI Overviews, rather than only by traditional search ranking. Also called AEO, answer engine optimisation.

Looking for how these work in practice? See the FAQ or pick your industry.

Start automating your customer calls today

Create your Leadvox AI account and launch your first voice calling agent in under 5 minutes with free trial credits.

Get Started for Free See Pricing

AI voice calling glossary

24terms you'll run into while evaluating AI calling software, explained in plain English — no vendor jargon, no assumed background.

AI voice agent: Software that makes or receives phone calls and holds a spoken conversation with the person on the other end. It listens, understands what was said, decides on a reply from its instructions and knowledge, and speaks that reply in a synthetic voice — all in under a second so the call feels natural.
Voice-to-voice latency: The delay between a caller finishing speaking and the agent starting to speak. It is the single most important quality measure in voice AI: humans take roughly 200ms to respond in conversation, so anything above about a second feels broken. Leadvox agents typically respond in 400ms to 800ms.
STT (speech-to-text): The component that converts the caller's audio into text so the AI can understand it. Also called ASR (automatic speech recognition). Accuracy on Indian languages, accents and code-mixed Hindi-English speech is what separates usable STT from unusable STT in this market.
TTS (text-to-speech): The component that converts the AI's text reply into spoken audio. Modern neural TTS produces speech with natural intonation and pacing rather than the flat, clipped delivery of older systems.
LLM (large language model): The model that decides what the agent says next, based on the conversation so far, the persona it was given, and any knowledge base documents attached to it. It is the reasoning layer between speech-to-text and text-to-speech.
Turn detection: Deciding when the caller has actually finished speaking rather than just pausing mid-sentence. Get it wrong in one direction and the agent interrupts; get it wrong in the other and it leaves an awkward silence. Also called endpointing.
Endpointing delay: How long the system waits after the caller stops making sound before treating the turn as finished. A short delay makes the agent fast but prone to interrupting; a long delay makes it patient but slow. Good systems vary this based on whether the sentence sounded complete.
Barge-in: The caller interrupting the agent mid-sentence, and the agent stopping to listen rather than talking over them. Without barge-in support, a call becomes a frustrating monologue.
Warm transfer: Handing a live call to a human while the caller stays on the line, so there is no callback and no repetition. The opposite is a cold transfer, where the caller is told someone will ring them back — which is where most leads are lost.
SIP trunk: The connection that carries phone calls between a software platform and the public telephone network over the internet. It is what allows an AI agent to dial a normal mobile number and be dialled from one.
DID (direct inward dialling) number: A normal phone number allocated to a business that routes incoming calls to a specific destination — in this context, to an AI agent rather than a desk phone.
Outbound calling: Calls the business makes to customers: new lead follow-up, renewal reminders, confirmation calls, win-back campaigns. Most AI calling value in India sits here, because outbound is the volume that human teams can never fully cover.
Inbound calling: Calls customers make to the business. An AI agent answering inbound calls gives consistent answers to routine questions and frees the front desk, transferring anything unusual to a person.
Lead response time: How long between a customer submitting an enquiry and someone actually calling them. It is the strongest single predictor of whether a comparison-shopped lead converts, and the reason instant auto-calling exists.
Lead ads: Ad formats on Facebook and Instagram where the user submits their contact details inside the app without visiting a website. They generate high volume cheaply, which makes fast, automated follow-up essential.
Lead qualification: Establishing whether a lead is actually worth a salesperson's time — budget, requirement, timeline, eligibility. Doing this on an automated first call means expensive human time is spent only on leads that pass.
Hot lead: A prospect whose conversation showed genuine, near-term buying intent. Leadvox labels each completed call so hot leads can be actioned first rather than found by listening to recordings.
Knowledge base: The documents you attach to an agent — brochures, price lists, policy documents, FAQ sheets — that it answers from during a call. It is what stops the agent inventing an answer, and updating one document updates every future call at once.
Agent persona: The plain-text instructions describing who the agent is, how it should speak, what it must ask, and what it must never do — such as never quoting a price or never giving medical advice.
Call transcript: The written record of everything said on a call by both parties. Far faster to review than audio, and what makes it practical to audit hundreds of calls rather than spot-checking three.
WhatsApp message template: A pre-approved message format that a business can send to someone who has not messaged it first. WhatsApp only reliably delivers business-initiated messages to new contacts as templates, which is why automated follow-up is built around them rather than free text.
RTO (return to origin): An e-commerce order that is shipped, refused or undelivered, and returned to the seller. The seller pays shipping both ways and earns nothing, which is why confirming cash-on-delivery orders by phone before dispatch matters so much in India.
COD confirmation call: A short call to a customer who placed a cash-on-delivery order, confirming that they placed it, that the address is right, and that they still want it — made before the parcel is dispatched, so a doubtful order can be cancelled instead of shipped twice.
GEO (generative engine optimisation): Making a website findable and quotable by AI answer engines such as ChatGPT, Claude, Perplexity and Google AI Overviews, rather than only by traditional search ranking. Also called AEO, answer engine optimisation.

Looking for how these work in practice? See the FAQ or pick your industry.

Start automating your customer calls today

Create your Leadvox AI account and launch your first voice calling agent in under 5 minutes with free trial credits.

Get Started for Free See Pricing