RexaJump to a category
Voice AI
10 terms
Voice agent
AI phone agent · Voice AI agent
A voice agent is software that conducts a spoken conversation with a person over a phone line or a browser connection, understanding natural speech and responding in speech.
Conversational AI
Conversational AI is the field of systems that interact with people through natural-language dialogue across text or speech.
Latency budget
Response latency · End-to-end latency
A latency budget is an allocation of the maximum acceptable end-to-end response delay across each component that contributes to it.
Time to first token
TTFT · Time to first byte
Time to first token is the delay between submitting a prompt to a language model and receiving the first unit of its output.
Interruption handling
Interruption handling is the policy that determines how a voice agent interprets and responds to a caller speaking over it.
Function calling
Tool use · Tool calling
Function calling is the mechanism by which a language model requests execution of a defined external operation and incorporates its result into the conversation.
System prompt
Agent instructions · System message
A system prompt is the standing instruction set given to a language model that defines its role, constraints and behaviour for every turn of a conversation.
Grounding
Grounding is the practice of constraining a model’s responses to information supplied from a verified source rather than its parametric knowledge.
Hallucination
Confabulation
A hallucination is model output that is fluent and confident but not supported by the model’s inputs or by fact.
Retrieval-augmented generation
RAG
Retrieval-augmented generation is an architecture in which relevant documents are fetched from a knowledge store at query time and supplied to a language model as context for its answer.
Speech
12 terms
Automatic speech recognition
ASR · STT · Speech-to-text
Automatic speech recognition is the conversion of spoken audio into written text by a machine.
Text-to-speech
TTS · Speech synthesis
Text-to-speech is the synthesis of natural-sounding spoken audio from written text.
Voice activity detection
VAD
Voice activity detection is the frame-by-frame classification of an audio stream into speech and non-speech.
Endpointing
Utterance segmentation · End-of-speech detection
Endpointing is the decision that a speaker has finished the utterance they were producing.
Turn detection
Turn-taking · End-of-turn detection
Turn detection is the decision that the floor has passed from one participant in a conversation to the other.
Barge-in
Barge-in is the ability of a caller to interrupt a system’s audio playback by speaking over it.
Speaker diarization
Diarisation · Speaker separation
Speaker diarization is the partitioning of an audio recording into segments labelled by who was speaking.
Word error rate
WER
Word error rate is the proportion of words a speech recogniser gets wrong, counting substitutions, deletions and insertions against a reference transcript.
Prosody
Prosody is the pattern of pitch, timing, loudness and rhythm layered over the words of an utterance.
SSML
Speech Synthesis Markup Language
SSML is an XML markup language for annotating text with instructions that control how a speech synthesiser renders it.
Voice cloning
Voice replication · Synthetic voice
Voice cloning is the creation of a synthetic voice that reproduces the vocal characteristics of a specific real person.
Speech-to-speech
S2S · Realtime speech model · Audio-native model
Speech-to-speech is an architecture in which a single model consumes audio and emits audio directly, without an intermediate text transcript driving the response.
Telephony
15 terms
PSTN
Public Switched Telephone Network · Landline network
The PSTN is the global interconnected system of circuit-switched and packet-switched telephone networks that carries ordinary phone calls.
SIP
Session Initiation Protocol
SIP is the signalling protocol used to establish, modify and terminate real-time voice and video sessions over IP networks.
SIP trunking
SIP trunk
SIP trunking is the delivery of telephone service to an organisation over an IP connection instead of physical phone lines.
DID
Direct Inward Dialing · Phone number
A DID is a telephone number assigned to an organisation that routes an inbound call directly to a specific destination without passing through an operator.
E.164
E164 · International number format
E.164 is the ITU standard that defines the international public telephone numbering plan, giving every number a unique globally routable form.
DTMF
Touch-tone · Dual-Tone Multi-Frequency
DTMF is the signalling system that encodes each telephone keypad press as a pair of simultaneous audio tones.
IVR
Interactive Voice Response · Phone menu
An IVR is an automated phone system that presents callers with a predefined menu and routes them according to keypad or spoken selections.
Call leg
Leg · Call segment
A call leg is a single connection between the telephony platform and one endpoint, of which a complete call may contain several.
Warm transfer
Attended transfer · Consultative transfer
A warm transfer is a handoff in which the transferring party briefs the receiving party before the caller is connected to them.
Cold transfer
Blind transfer · Unattended transfer
A cold transfer is a handoff in which the caller is sent to another destination immediately, with no consultation beforehand.
Carrier
Telecom operator · Voice provider
A carrier is a company that operates telephone network infrastructure and terminates or originates calls on it.
CNAM
Caller ID Name · Caller name display
CNAM is the caller-name information displayed alongside a calling number on the recipient’s handset.
STIR/SHAKEN
STIR · SHAKEN · Call authentication
STIR/SHAKEN is a framework of standards by which carriers cryptographically sign outbound calls to attest to the legitimacy of the calling number.
A2P 10DLC
10DLC · Application-to-Person 10-digit long code
A2P 10DLC is the US registration framework under which businesses declare their identity and messaging use case before sending application-to-person traffic from ordinary ten-digit numbers.
Concurrency
Concurrent calls · Channel capacity
Concurrency is the number of calls a system is handling simultaneously at a given instant.
Compliance
9 terms
TCPA
Telephone Consumer Protection Act
The TCPA is a US federal statute restricting telemarketing calls, automated dialling, prerecorded and artificial voice messages, and unsolicited texts.
Do Not Call
DNC · Suppression list
Do Not Call refers to the registries and internal lists of numbers that must be suppressed from telemarketing contact.
Calling window
Quiet hours · Permitted calling hours
A calling window is the range of local hours during which it is permissible to place a call to a recipient, with the hours outside it known as quiet hours.
Two-party consent
All-party consent
Two-party consent is the requirement in certain jurisdictions that every participant in a conversation agree before it may lawfully be recorded.
Recording consent
Call recording disclosure
Recording consent is the permission obtained from call participants before their conversation is captured and stored.
Opt-out keyword
Stop keyword · Revocation phrase
An opt-out keyword is a word or phrase that, when spoken or sent by a recipient, is treated as a request to stop being contacted.
Consent capture
Consent capture is the process of obtaining, recording and retaining evidence that a person agreed to be contacted.
Audit trail
Audit log
An audit trail is an append-only record of significant actions in a system, capturing who did what, when, and to which resource.
GDPR
General Data Protection Regulation
The GDPR is the European Union regulation governing the processing of personal data relating to people in the EU and EEA.
Operations
7 terms
WebRTC
Web Real-Time Communication
WebRTC is a browser and mobile standard for real-time audio, video and data communication directly between endpoints without a plugin.
Webhook
HTTP callback · Event notification
A webhook is an HTTP request sent by a service to a URL you control when an event occurs, delivering the event rather than waiting to be polled.
Idempotency key
An idempotency key is a client-supplied identifier that lets a server recognise a retried request and return the original result instead of performing the operation again.
Structured outcome
Disposition · Call outcome
A structured outcome is a machine-readable classification of what a call achieved, drawn from a defined set of possible results.
Credit ledger
Usage ledger · Balance ledger
A credit ledger is an append-only record of the balance changes in a prepaid usage account, from which the current balance is derived.
Connect rate
Answer rate · Contact rate
Connect rate is the proportion of dialled calls that reach a live person, as distinct from those that go unanswered or reach voicemail.
Abandonment rate
Abandoned call rate
Abandonment rate is the proportion of connected calls that end without the recipient reaching the party or agent they were meant to reach.