Google adds streaming output to Gemini 3.1 Flash TTS, enabling real-time speech generation via streamGenerateContent

Google on June 17 enabled streaming support for the gemini-3.1-flash-tts-preview model, allowing developers to receive speech audio as it is generated rather than waiting for a complete synthesis pass. The update works through both the standard streamGenerateContent API and the stream: true flag in the Interactions API.

What's new

Previously, text-to-speech requests to the Gemini API required a full round-trip — the model would synthesize the complete audio response before returning anything to the client. With the June 17 update, streaming via streamGenerateContent is now supported for gemini-3.1-flash-tts-preview, delivering audio chunks as they are produced. This is the same streaming paradigm developers use for text generation, now extended to speech output.

The stream: true parameter in the Interactions API (Google's conversational interface layer) is also supported, enabling real-time dialogue applications where speech latency is a critical factor.

Google's Text-to-Speech guide in the Gemini developer documentation covers the implementation details for both paths.

Context

The gemini-3.1-flash-tts-preview model was introduced as part of Google's push to bring multimodal output — including natively generated speech — into the Gemini API. The preview designation indicates the model is still in active development, but the streaming capability significantly improves its suitability for latency-sensitive production use cases.

Streaming text-to-speech is a well-established requirement for conversational voice interfaces: without it, even a fast synthesis model introduces noticeable lag because the user hears nothing until the full response is ready. Most production voice applications (phone IVR, voice assistants, live agent call assistance) stream audio from the first chunk.

The June 17 update aligns Gemini's TTS behavior with its text generation streaming, which developers have had access to since the API's earliest days.

Why it matters

For developers building voice-first applications on the Gemini API — live customer support agents, real-time translation, interactive voice response systems — streaming TTS removes the main architectural objection to using the Gemini-native speech stack. Previously, a team building a low-latency voice product would need to route through a separate TTS provider to avoid the full-synthesis wait; that workaround is now unnecessary for applications already running on Gemini.

The update is also a signal about where Google is investing in the Gemini API: multimodal output (text, images, audio) with streaming parity is becoming a platform standard, not a premium feature.