Voice AI Glossary
Plain-English definitions of the voice AI, telephony, and call-operations terms you meet when evaluating or building AI phone agents.
10DLC10DLC is the U.S. carrier framework for registered application-to-person text messaging sent through standard local numbers for business use cases.A2P messagingA2P messaging is text sent from a software application to a person's phone for a defined business purpose, such as a confirmation or service notice.A/B testing for voice agentsA/B testing for voice agents compares variants on similar live calls to measure which version performs better against a defined outcome.Abandonment rateAbandonment rate is the share of eligible inbound callers who disconnect before a human or automated agent begins handling their request for service.After-hours coverageAfter-hours coverage handles calls outside normal staffed hours through automation, alternate routing, on-call escalation, or message intake for later action.AI answering serviceAn AI answering service receives business calls, handles approved requests, records caller details, and escalates calls under defined rules.AI phone agentAn AI phone agent handles inbound or outbound calls by understanding speech, responding in real time, and carrying out defined phone workflows.AI receptionistAn AI receptionist answers business calls, gathers caller information, handles routine requests, and routes conversations by configured rules.Answering machine detection (AMD)Answering machine detection classifies whether an answered outbound call reached a person or an automated greeting, then routes the call accordingly.Appointment remindersAppointment reminders are scheduled messages or calls that restate booking details and ask the recipient to confirm, cancel, or reschedule.Appointment settingAppointment setting is the process of finding an eligible time, collecting required details, and creating a confirmed booking for a caller.Audio codecAn audio codec is a method for encoding sound into digital data for transmission or storage and decoding it back into playable audio at the receiving endpoint.Automatic call distribution (ACD)Automatic call distribution is a routing system that assigns inbound calls to suitable agents or resources according to rules and availability.Automatic speech recognition (ASR)Automatic speech recognition converts spoken audio into machine-readable words so software can interpret, search, transcribe, or act on speech.Average handle time (AHT)Average handle time measures the average time needed to complete a handled interaction, including the conversation, holds, and defined follow-up work.Average speed of answer (ASA)Average speed of answer measures how long answered inbound callers wait in a queue before a human or automated agent begins handling their request.BackchannelingBackchanneling uses brief listener responses such as acknowledgements to show attention without taking over the speaker's conversational turn.Barge-inBarge-in is the ability for a caller to interrupt a voice agent while it is speaking so the system stops playback and listens to the caller.Bot disclosureBot disclosure is a clear notice that tells a caller they are interacting with an automated or AI system rather than a human during a phone call.Branded callingBranded calling is a caller-identification service that can show a verified business name, logo, or call reason on supported phones and networks.Bring your own carrier (BYOC)Bring your own carrier lets a business connect its existing telephony provider or SIP trunk to another platform while retaining carrier control.Business Associate Agreement (BAA)A Business Associate Agreement is a HIPAA contract governing how a business associate may handle protected health information on another entity's behalf.Call analyticsCall analytics turns call metadata, conversation records, and outcomes into measures that help teams understand performance and caller needs.Call deflectionCall deflection redirects or resolves an inbound request through self-service or another appropriate path before a staffed phone conversation is needed.Call dispositionA call disposition is a structured outcome label assigned after a call to support consistent reporting and determine the appropriate next workflow action.Call flowA call flow is the ordered sequence of prompts, decisions, actions, and fallback paths that takes a phone call from initial connection through completion.Call forwardingCall forwarding redirects an incoming call from its original number or endpoint to another destination under an always-on or conditional routing rule.Call monitoringCall monitoring is the live or near-live observation of an active phone conversation for operational oversight, timely support, or intervention when needed.Call quality assurance (QA)Call quality assurance is the systematic review of conversations against defined standards to find risks, coach performance, and improve call flows.Call queueA call queue holds and orders inbound calls until an appropriate agent or other answering resource becomes available to handle each caller.Call recording consentCall recording consent is the notice, permission, or other lawful basis required before a telephone conversation may be recorded during a call.Call recordingCall recording captures and stores the audio of a phone conversation for later playback, quality review, documentation, training, or dispute resolution.Call routingCall routing is the process of selecting where a call goes next based on rules, caller information, business context, and destination availability.Call scoringCall scoring evaluates a conversation against a defined rubric so teams can compare outcomes, behavior, accuracy, and process adherence consistently across calls.Call screeningCall screening evaluates an incoming call before connecting it, using caller information and purpose to choose whether, where, or how it should proceed.Call simulationCall simulation is a controlled test that uses realistic calls to evaluate how a voice agent handles defined scenarios before broader use.Call summaryA call summary is a concise post-call account of the conversation's purpose, key facts, decisions, outcome, and remaining follow-up for review or handoff.Call transcriptionCall transcription converts spoken audio in a phone conversation into a time-ordered text record for search, review, analysis, and downstream processing.Call whisperCall whisper is a private audio message delivered to one call participant before or during a conversation without being heard by the other participant.Caller IDCaller ID is the calling-party information presented to the recipient or receiving network, usually a phone number and sometimes a display name.CNAMCNAM is caller-name data associated with a North American phone number and shown by participating networks alongside the number on inbound calls.Code-switchingCode-switching is the shift between languages or language varieties within one conversation, sometimes within a single sentence or speaking turn.Cold transferA cold transfer sends an active caller directly to another person or endpoint without first briefing the recipient or confirming that someone will answer.Computer telephony integration (CTI)Computer telephony integration connects a phone system with business software so applications can use call events, controls, and caller context.Concurrent calls (concurrency)Concurrent calls are calls active at the same time, and concurrency is the number of overlapping calls a system must handle without delaying or rejecting new callers.Consent captureConsent capture records a person's affirmative permission for defined communications together with evidence of what they agreed to and when.Contact centerA contact center is the people, processes, and technology a business uses to manage customer conversations across phone and digital channels.Containment rateContainment rate is the share of eligible automated interactions completed without a transfer to a human agent or other live human intervention during the call.Context windowA context window is the limited amount of text a language model can consider at once, including instructions, dialogue, retrieved content, and tool results.Conversational AIConversational AI enables software to understand, manage, and respond to human language across multi-turn interactions in voice or text channels.CRM integrationA CRM integration connects phone workflows with customer records so authorized systems can retrieve context and write back call outcomes and next steps.CSAT (Customer Satisfaction Score)CSAT is a post-interaction measure of how satisfied customers say they are with a specific call, outcome, product, or service experience they received.Customer Effort Score (CES)Customer Effort Score measures how easy or difficult customers say it was to complete a task, resolve an issue, or get an answer during a recent interaction.Data Processing Agreement (DPA)A Data Processing Agreement sets the rules for a processor handling personal data on a controller's behalf and allocates key privacy duties.Data residencyData residency describes the geographic location where an organization’s information is stored or processed by a service and its providers.Data retentionData retention is the policy and practice of keeping information for a defined period before deleting, anonymizing, or archiving it.Dialogue managementDialogue management determines what a conversational system should do next based on the caller's goal, known information, and call state.DID number (Direct Inward Dialing)A DID number is a public telephone number routed directly to a specific user, team, extension, application, or location in a business phone system.Do Not Call (DNC)Do Not Call describes registries and business-specific requests used to prevent covered telemarketing calls to people who do not want them.DTMF (touch-tone signaling)DTMF is the touch-tone signaling method that sends paired audio tones from a phone keypad to communicate digits and symbols within a live call.E.164 phone number formatE.164 is the international phone-number format that represents a dialable number with a country code and national number digits for consistent routing.Embeddable voice widgetAn embeddable voice widget lets website visitors start a spoken conversation with a voice agent from a web page without dialing a phone number.EndpointingEndpointing is the decision process a voice system uses to determine that a caller has finished a spoken turn and that it is time to respond.Entity extractionEntity extraction finds and structures specific details in a caller's language, such as names, dates, locations, quantities, and identifiers.EscalationEscalation moves a call to a person, specialist, or higher-authority process when the current agent cannot handle it safely, appropriately, or completely.First call resolution (FCR)First call resolution measures whether a caller's issue is fully resolved during the initial call without requiring another contact or follow-up.Function calling (tool use)Function calling lets an AI agent request a defined external action with structured inputs, then use the result to continue the conversation.GDPRThe GDPR is the European Union data protection law governing how organizations collect, use, secure, and delete personal data and the rights individuals hold over it.GroundingGrounding ties an AI phone agent's response to supplied evidence, such as approved documents, call context, or results returned by business tools.GuardrailsGuardrails are instructions and controls that limit what an AI phone agent may say or do and define how it handles risky or uncertain situations.HallucinationA hallucination is an AI-generated statement presented as factual even though the information available to the system during a call does not support it.HIPAAHIPAA is the U.S. federal framework for how covered entities and business associates use, disclose, and protect identifiable health information.HMAC signatureAn HMAC signature uses a shared secret and a hash function to let a recipient verify that a message is authentic and unchanged.Human handoffA human handoff moves an active conversation from an automated agent to a person when the caller's request needs human attention, judgment, or authority.Human in the loopHuman in the loop is an operating model in which people review, guide, approve, or take over selected decisions and actions within an automated process.Intent detectionIntent detection identifies what a caller is trying to accomplish so a voice agent can choose the appropriate response, workflow, or handoff.IVR (Interactive Voice Response)IVR is an automated phone system that collects keypad or spoken input to provide information, perform actions, or route calls without an operator.JitterJitter is variation in packet arrival timing that can make a live voice call sound choppy, delayed, or uneven even when the connection stays active.Knowledge baseA knowledge base is a curated set of documents and data that an AI phone agent can search to answer questions with relevant business information.Large language model (LLM)A large language model is a statistical model trained on extensive text data to interpret instructions, generate language, and perform text-based tasks.Latency in voice AILatency in voice AI is the delay between a caller's speech or action and the system's audible response, including processing and delivery time.Lead qualificationLead qualification is the structured process of deciding whether a prospect fits defined criteria and should advance to a specific next step.Letter of Authorization (LOA)A Letter of Authorization permits a carrier or provider to act on a customer's behalf for a defined number-porting or telephone-service request.Mean Opinion Score (MOS)Mean Opinion Score is a summary rating of perceived voice quality, based on listener judgments or a model designed to estimate those judgments.Missed-call recoveryMissed-call recovery is the process of following up after an unanswered inbound call so the caller can resume the intended conversation or request.Model Context Protocol (MCP)Model Context Protocol is a standard interface for AI applications to discover and invoke tools or retrieve context exposed by connected servers.Multi-turn conversationA multi-turn conversation is an exchange in which later questions, answers, or actions depend on information established in earlier turns.Multilingual voice agentA multilingual voice agent can understand and respond in more than one configured language while carrying context through a spoken conversation.Natural language processing (NLP)Natural language processing is the field of computational methods used to analyze, transform, understand, and generate human language in software systems.Natural language understanding (NLU)Natural language understanding derives machine-usable meaning from language, such as a speaker's intent, referenced entities, and conversational context.Net Promoter Score (NPS)Net Promoter Score measures customers' stated willingness to recommend a company by comparing the shares of promoters and detractors in a defined survey.Neural text-to-speechNeural text-to-speech uses learned neural networks to generate spoken audio from text, including its pronunciation, timing, and vocal expression.No-show recoveryNo-show recovery is the follow-up process used after a person misses a scheduled appointment to resolve its status and arrange a next step.Number portingNumber porting is the carrier process for moving a telephone number to a new service provider while keeping the public number customers dial unchanged.OAuthOAuth is an authorization framework that lets an application receive limited access to another service without obtaining a user’s password.OmnichannelOmnichannel service connects customer identity, context, routing, and history across communication channels so each interaction can continue the journey.Opt-out managementOpt-out management is the process of receiving, recording, and enforcing a person's request to stop defined calls or other communications.Outbound campaignAn outbound campaign is a coordinated program that places calls to an eligible contact list using defined timing, conversation, and outcome rules.Overflow answeringOverflow answering routes calls to backup capacity when the primary team cannot answer within the business's chosen conditions or time threshold.Packet lossPacket loss occurs when some audio data sent across a network never reaches its destination, which can create gaps, distortion, or missing speech.PBXA PBX is a private business phone system that connects internal extensions and controls how calls move between users, outside lines, and locations.Personally identifiable information (PII)Personally identifiable information is data that identifies a specific person directly or can be linked with other data to do so in a given context.PhonemeA phoneme is the smallest sound category that can distinguish one word from another in a language, regardless of how the sound is spelled.Power dialerA power dialer calls contacts sequentially when capacity is available, reducing manual dialing while keeping outreach tied to an ordered contact list.Predictive dialerA predictive dialer places outbound calls ahead of agent availability and uses pacing estimates to connect live answers to available agents.Prompt engineeringPrompt engineering is the disciplined process of writing, testing, and refining instructions so an AI agent behaves reliably across real scenarios.Pronunciation dictionaryA pronunciation dictionary maps written terms to approved spoken forms so speech systems handle names, acronyms, and specialized vocabulary consistently.ProsodyProsody is the pattern of rhythm, stress, pitch, loudness, and timing that carries structure and meaning beyond the words in spoken language.Protected health information (PHI)Protected health information is identifiable health data that a covered entity or business associate holds or transmits in a HIPAA-covered context.PSTNThe PSTN is the public telephone network that assigns dialable numbers and routes calls among fixed, mobile, and business phone systems across carriers.Quiet hoursQuiet hours are protected time windows when an outbound calling system does not contact recipients based on local time, law, or business policy.RedactionRedaction is the removal or masking of sensitive information so it is no longer visible or available in a record that is viewed, shared, or retained.Regression testing for voice agentsRegression testing for voice agents reruns consistent call scenarios to detect behavior changes after prompts, tools, or workflows are updated.REST APIA REST API is an HTTP interface that organizes operations around resources and uses standard methods to create, read, update, or delete them.Retrieval-augmented generation (RAG)Retrieval-augmented generation finds relevant source material at response time and gives it to a language model to produce a grounded answer.Retry policyA retry policy defines whether, when, and how often another call attempt may occur after a specific unsuccessful or incomplete outcome.RTP (Real-time Transport Protocol)RTP is a network protocol that carries timestamped, sequenced audio or video packets so receivers can play real-time media in order during an internet call.Screen popA screen pop automatically displays relevant caller and interaction context in an agent's workspace when a call arrives, is answered, or is transferred.Sentiment analysisSentiment analysis estimates a caller's emotional tone from spoken or transcribed language to help teams interpret and review interactions.Service level agreement (SLA)A service level agreement defines a provider's measurable service commitments, how performance is calculated, and what happens when a commitment is missed.SIP REFERSIP REFER is a signaling method that asks an endpoint to initiate a new SIP request to a specified target, commonly to transfer a call between parties.SIP trunkA SIP trunk is an IP-based connection that links a phone system or voice application to a telephony provider for call routing and number delivery.SIP URIA SIP URI is an address that identifies a user, service, or endpoint for routing and delivering a SIP session across an IP communications network.SIP (Session Initiation Protocol)SIP is a signaling protocol used to start, modify, route, and end real-time voice or video communication sessions, including phone calls, over IP.Slot fillingSlot filling is the process of collecting and validating the specific pieces of information required to complete a conversational task."Spam Likely" labeling"Spam Likely" labeling is a carrier or device warning that a call may be unwanted, based on reputation signals rather than a universal verdict.Speaker diarizationSpeaker diarization separates a call recording into speaker-labeled segments, showing who spoke when and supporting clearer transcripts and review.Speech-to-text (STT)Speech-to-text converts spoken audio into written words for transcripts, search, analysis, accessibility, conversational software, and call workflows.Speed to leadSpeed to lead measures the time between a prospect's qualifying action and the business's first defined response to that prospective customer.SRTPSRTP is a security profile for real-time media that encrypts RTP payloads and helps detect tampering and replayed packets during live voice or video sessions.SSML (Speech Synthesis Markup Language)SSML is a markup language that gives a compatible speech synthesizer instructions for pronunciation, pauses, emphasis, pacing, and pitch.STIR/SHAKENSTIR/SHAKEN is a caller-ID authentication framework that lets phone networks assess an originating provider's authority to use a calling number.Suppression listA suppression list is a controlled set of phone numbers or contacts that a dialing workflow must exclude before an outbound call or retry is placed.System promptA system prompt is the instruction set that defines an AI agent's role, priorities, boundaries, and rules before a caller's conversation begins.TCPA (Telephone Consumer Protection Act)The TCPA is a U.S. federal law that restricts certain calls, texts, and faxes and establishes protections for consent and calling preferences.Text-to-speech (TTS)Text-to-speech converts written text into spoken audio, giving a voice application a consistent way to deliver generated responses aloud on a live call.Time to first audioTime to first audio measures how long a voice system takes to begin playing an audible response after a defined input event in a live interaction.Toll-free numberA toll-free number lets callers reach an organization without a separate long-distance charge, while the recipient pays for the inbound phone service.Turn-takingTurn-taking is the coordination of who speaks, who listens, and when conversational control passes between a caller and a voice agent during a call.Two-party consentTwo-party consent is a recording rule that requires every participant's permission before certain conversations on a call may lawfully be recorded.Virtual receptionistA virtual receptionist performs front-desk call handling from outside the physical office through human operators, software, or a hybrid service.Voice activity detection (VAD)Voice activity detection identifies which parts of an audio stream contain human speech so a voice agent can listen and respond at the right time.Voice agentA voice agent is software that conducts spoken conversations in real time, follows task-specific instructions, and can take actions for the caller.Voice AIVoice AI combines speech recognition, language processing, and speech generation so software can understand spoken input and respond aloud in real time.Voice cloningVoice cloning creates a synthetic voice that resembles a particular speaker and can produce new speech the person did not originally record.Voice user interface (VUI)A voice user interface lets a person operate a system through spoken input and audio responses instead of relying only on screens or controls.Voicemail detectionVoicemail detection identifies when an outbound call has reached a mailbox so the system can leave a message, hang up, or schedule another attempt.Voicemail dropVoicemail drop leaves a prepared message after an outbound call reaches voicemail, then records the result without starting a live conversation.VoIPVoIP carries voice calls as digital data over internet protocol networks instead of relying entirely on dedicated circuits for call routing.Warm transferA warm transfer privately briefs the intended recipient, giving that person context and a chance to accept before connecting the caller to a live conversation.WebhookA webhook is an automated HTTP message sent when an event occurs, allowing another system to react without repeatedly checking for updates.WebRTCWebRTC is a set of browser and application technologies for sending real-time audio, video, and data over an internet connection with low delay.WebSocketWebSocket is a protocol that maintains a persistent, two-way connection so a client and server can send messages without new HTTP requests.Zero data retentionZero data retention is an arrangement in which a service does not keep covered customer content after it finishes the required processing.