Sub-second latency

Sub-second latency is the end-to-end response time under one second between a user speaking and an avatar replying, covering speech recognition, model reasoning, speech synthesis, animation, and network streaming.

How sub-second latency works

Sub-second latency means the avatar responds in less than one second from the user's turn to the avatar's reply. In a real-time avatar system, that timing includes audio capture, speech recognition, model reasoning, text-to-speech, face animation and network delivery.

The hard part is that every stage takes a slice of the budget. If ASR is slow, or the LLM waits too long before producing tokens, or TTS only starts after the full answer is complete, the conversation starts to feel delayed.

A concrete example: a user asks a checkout assistant whether a discount applies. The system has to recognise the question, query the right policy, speak back and animate the face quickly enough that the exchange still feels live.

Under one second is the minimum bar for conversational feel. The closer the system gets to a few hundred milliseconds, the more natural turn-taking, interruption and back-and-forth become.

Response time

User experience

What Anam ships

Anam's Cara-4 model delivers expressive real-time avatars with around 150 ms server-side avatar-generation latency once a session is running, across 70+ languages. Builders use JavaScript and Python SDKs or integrations for LiveKit, Pipecat, ElevenLabs Agents, Agora, and VideoSDK. Bring any AI stack including OpenAI, Claude, Gemini, Mistral, Groq, Deepgram, Cartesia, or custom providers. The platform supports WebRTC delivery, SOC 2 Type II, HIPAA, zero data retention, and regional data residency. Sessions stream low-latency audio and video to browsers and native apps.

Explore the real-time avatar API →

Related terms

Latency budget

A latency budget is the total response-time allowance split across speech recognition, model reasoning, tools, text-to-speech, rendering, and streaming so a real-time avatar can stay conversational.

WebRTC

WebRTC is the browser standard for low-latency live audio and video streaming, commonly used to carry microphone input and avatar output during real-time interactive sessions.

Turn detection

Turn detection is the process of deciding when a user has finished speaking, so the avatar can respond at the right moment without interrupting or waiting too long.

TTS

TTS, or text-to-speech, is the system that converts a model's text response into spoken audio, giving an avatar a voice that can be streamed and lip-synced in real time.

Frequently asked questions

Why does sub-second latency matter for avatars?

Users notice pauses quickly in spoken conversation. Sub-second latency keeps the avatar feeling responsive enough for natural turn-taking instead of delayed playback.

What counts toward avatar latency?

The full loop counts: microphone capture, speech recognition, model reasoning, tool calls, text-to-speech, face animation, network delivery, and playback in the client.

Is under one second always good enough?

Under one second is a useful baseline, but faster is better. Around 250ms can feel much more human, while 800ms or more starts to feel broken.

How do teams reduce avatar latency?

Teams reduce latency with streaming ASR, fast first-token model output, low-startup TTS, efficient face animation, WebRTC delivery, and careful control over tool-call delays.

Last updated: 17th July 2026 · Reviewed quarterly.

← Back to glossary