Streaming STT

Streaming STT (speech-to-text) emits partial transcriptions as the user speaks — instead of waiting for end-of-utterance — enabling sub-second response from a voice assistant.

Production voice agents in 2026 require streaming STT. Models like Deepgram Nova-3, AssemblyAI Universal-Streaming, and OpenAI Whisper-3 emit partial transcripts every ~100-200ms. The agent can begin the LLM call before the user finishes speaking (if VAD signals end-of-utterance with high confidence) or use the streaming text to begin retrieval / classification. End-to-end voice agent latency targets in 2026 are <800ms from end-of-speech to start-of-agent-speech; without streaming STT this is unachievable.

When to use streaming stt

• Realtime voice agents.
• Live captioning.
• Interactive voice UX.

Common mistakes

• Using batch STT for realtime use — instant latency penalty.
• Ignoring word-error rate at the cost of latency — quality matters too.

FAQ

Related terms

• Voice (LLM apps) — Voice in LLM apps refers to the full speech pipeline — speech-to-text (STT), language model, text-to-speech (TTS) — that lets users converse with an AI assistant in spoken language.
• AI agent — An AI agent is a system where a language model autonomously plans and executes a sequence of tool calls to accomplish a goal.
• Response streaming — Response streaming pipes the model's output token-by-token to the client as it's generated, so users see text appearing in real time instead of waiting for the full answer.

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/streaming-stt.md.