Streaming Text-to-Speech: Hear Audio as It Generates

What Is Streaming TTS?

With traditional (async) TTS, you submit text and wait for the entire audio file to be generated before you can play it. Streaming TTS flips this — audio starts playing within milliseconds of generation beginning.

The provider generates audio in chunks and sends them to your client as they're ready. You hear the first words while the last words are still being synthesized.

When Streaming Makes Sense

• Interactive applications where latency matters (chatbots, voice assistants)
• Long content where waiting 10+ seconds for a full file feels slow
• Preview workflows where you want to hear the voice before committing
• Real-time narration synced to user actions

When Async Is Better

• Batch processing hundreds of files overnight
• Archival where you need a perfect, complete file
• Formats that need post-processing (custom bitrate, sample rate conversion)

How It Works Under the Hood

1. Your client sends text to the TTS API with delivery_mode: "stream"
2. The API returns a one-time stream_url
3. Your client fetches that URL — audio bytes flow immediately
4. Meanwhile, the backend saves the complete file for later access
5. The durable file is available at audio_endpoint once generation completes

The key insight: streaming is a delivery mode, not a durability mode. Your audio is still saved permanently — you just get to hear it sooner.

Try It

AI TTS Microservice supports streaming for Google Chirp3HD, Polly, and Kokoro voices. Select "Stream" delivery in the UI or pass delivery_mode: "stream" in the API.

Try streaming →