The real-time AI
avatar glossary.

Acoustic echo cancellation removes the avatar’s speaker output from the microphone signal using a playback reference.

Audio backpressure is flow control applied when a producer generates audio faster than downstream components can consume it.

Audio chunking divides a continuous speech stream into ordered blocks that can be processed incrementally.

Audio prebuffering accumulates a minimum amount of media before playback or downstream processing begins.

Audio resampling converts a signal from one sample rate to another while preserving its perceived timing and content.

Sample rate is the number of audio samples captured or represented per second, measured in hertz.

Audio-only fallback continues speech playback when avatar animation or rendering is unavailable.

Audio-video synchronization aligns avatar motion with the corresponding moments in speech playback.

An avatar asset bundle packages the meshes, textures, rig data, animations, and metadata needed to render an avatar.

Avatar asset preloading fetches and prepares required resources before the avatar must first appear or speak.

Avatar frame rate is the number of visual frames actually presented per second.

An avatar motion frame is a timestamped set of pose or facial-control values representing one instant of animation.

An avatar render loop repeatedly updates animation state and submits the next visual frame for display.

An avatar session is a bounded runtime context containing configuration, authentication, connection, and conversational state.

Avatar speech audio is the speech signal the avatar should perform, normally the output of a TTS system rather than the user’s microphone.

A bandwidth budget is the planned sustained and peak network allowance for one avatar session.

Barge-in allows a user to interrupt an avatar’s active response by speaking.

A buffer underrun occurs when playback consumes all ready media before the next required data arrives.

Coarticulation is the way neighboring speech sounds influence the mouth movement used to produce each sound.

Connection state is an explicit representation of a runtime connection’s current lifecycle status.

A conversation ID is a stable identifier for one multi-turn interaction between a user and an avatar.

Conversational overlap occurs when the user and avatar speak at the same time.

A dropped frame is a scheduled visual frame that is not presented because it arrived too late or could not be rendered.

An end-of-input signal marks that no more audio chunks belong to the current avatar utterance.

End-of-turn detection estimates when a user has finished a conversational turn rather than merely paused.

A facial blendshape is a predefined mesh deformation combined with other shapes through numeric weights.

Frame pacing describes how evenly rendered frames are presented over time.

Interruption latency is the time between detectable user interruption and the avatar’s speech and motion actually stopping.

A jitter buffer temporarily holds incoming media to absorb variation in packet arrival timing.

A keepalive heartbeat is a periodic liveness signal used to detect a connection that has silently stopped carrying traffic.

Lip-sync drift is a timing error between speech audio and mouth movement that changes or accumulates during playback.

A motion inference window is a buffered interval of speech analyzed together to generate the next segment of avatar movement.

P95 latency is the value at or below which 95 percent of measured latency observations fall.

PCM16 is uncompressed linear audio represented as signed 16-bit samples.

Phoneme-to-viseme mapping converts linguistic speech-sound labels into visible mouth-shape categories.

Real-time factor is processing time divided by the duration of the media produced or analyzed.

Reconnect backoff progressively delays retry attempts after repeated connection failures.

Semantic endpointing uses linguistic context to predict whether a speaker’s thought is complete.

A session token is a short-lived credential that authorizes an avatar client without exposing a permanent backend API key.

Shader warmup compiles and links graphics programs before their first visible use.

Speech-driven animation generates timed facial or body movement from features in a spoken audio signal.

Time to first audio is the interval from a defined request boundary to the first audible sample of the avatar’s response.

Time to first motion is the interval from a declared speech-input boundary to the first usable or visible avatar motion.

Token expiration is the time after which a session credential is no longer accepted.

TTS generation speed is how quickly synthesized audio is produced relative to the duration of the resulting speech.

Turn-taking is the control logic that decides whether the user or avatar currently holds the conversational floor.

A viseme is a visually distinguishable mouth shape associated with one or more speech sounds.

Voice activity detection classifies short audio frames as human speech or non-speech.

WebAssembly avatar rendering uses a compiled browser module for performance-sensitive runtime logic alongside a web graphics API.

WebGL context loss occurs when the browser invalidates a page’s GPU rendering state and resources.