realtime text-to-speech

Realtime text-to-speech with the top-ranked voice, fast enough for live conversation.

Simba 3.2 is streaming-native, sub-300ms first byte
Statistically tied for #1 on Artificial Analysis
Powers realtime voice agents and live captions

definition

Realtime text-to-speech is speech generation fast enough for a live app, where someone is waiting to hear the answer. The metric that matters is time-to-first-byte: how quickly the first audio arrives. Everything else is downstream of that number.

simba-3-2

Simba 3.2 is built for this. It is streaming-native, produces first byte in under 300ms, and is statistically tied for first place on the Artificial Analysis leaderboard at roughly a third of the price of the model it ties with. For a live application, that combination of speed, quality, and price is the whole game.

streaming-architecture

Realtime playback comes from the streaming endpoint. POST /v1/audio/stream returns audio in chunks over a single HTTP connection, so playback starts on the first chunk instead of after the full clip. Paired with a streaming-native model, the perceived delay stays under the threshold where a listener notices it.

live-app-patterns

Voice agents, live captioning, live translation, in-game dialogue, and interactive stories all depend on realtime TTS. Each needs audio that starts fast and keeps flowing, which is exactly what the streaming endpoint plus Simba 3.2 delivers.

code-samples

curl -X POST https://api.speechify.ai/v1/audio/stream \
  -H "Authorization: Bearer $SPEECHIFY_API_KEY" \
  -H "Accept: audio/mpeg" \
  -H "Content-Type: application/json" \
  -d '{"input": "Streaming in real time.", "voice_id": "george", "model": "simba-3.2"}'

FAQ

Frequently asked questions

What is realtime text-to-speech?

Realtime text-to-speech generates audio fast enough to use in a live application, where a person is waiting to hear the reply. The key metric is time-to-first-byte: how quickly audio starts. Simba 3.2 is streaming-native with sub-300ms first byte, which keeps live conversations feeling immediate.

Is it fast enough for a voice agent?

Yes. Text-to-speech first-byte time is the last stage before a caller hears a reply, and Simba 3.2's sub-300ms start leaves room for transcription and reasoning inside the per-turn budget. It is the TTS behind SpeechifyAI's own voice agents.

How do I get the lowest latency?

Use the streaming endpoint with model simba-3.2, and start playback as the first chunks arrive rather than waiting for the full clip. Streaming plus a streaming-native model is what keeps perceived latency low.

Text-to-Speech API streaming text-to-speech Voice Agents Voice Agent API Cartesia text to speech latency

Start building

Realtime text-to-speech with the top-ranked voice, fast enough for live conversation.

Get an API key, free Talk to sales