HTTP Streaming

POST /v1/tts/stream synthesizes the full input text and streams the audio bytes back as they are generated, over one HTTP response. This lowers time-to-first-byte compared to waiting for the complete file.

The request body is identical to POST /v1/tts/generate. Authentication, rate limits, credits, voice/language, and model_id are all validated before streaming starts.

pcm is the recommended output_format for streaming since it has no container overhead and is ready for custom playback pipelines.

Response

The response body is application/octet-stream. Usage metadata is returned in headers:

12345X-RateLimit-Limit:        15
X-RateLimit-Remaining:    13
X-RateLimit-Reset:        1715678460
X-Characters-Used:        12
X-Audio-Duration-Seconds: 1.8

Streaming to a file

123456789with open("speech.pcm", "wb") as file:
    for chunk in client.tts.stream(
        text="வணக்கம், நான் வாக்யம் AI பேசுகிறேன்.",
        model_id="raaga-v1",
        voice="Archana",
        language="ta-IN",
        output_format="pcm",
    ):
        file.write(chunk)

Collecting the full stream with metadata (Python)

If you want the complete audio plus the response metadata in one call, the Python SDK offers stream_to_bytes:

Python

12345678910streamed = client.tts.stream_to_bytes(
    text="வணக்கம்.",
    model_id="raaga-v1",
    voice="Archana",
    language="ta-IN",
)

streamed.save("speech.pcm")
print(streamed.metadata.characters_used)
print(streamed.metadata.duration_seconds)

When to use streaming vs. realtime

HTTP streaming is best when you have the full text up front and want the audio to start playing sooner.
WebSocket is best when you produce text sentence-by-sentence (for example from an LLM) and want each sentence synthesized as it arrives. See the realtime guide.