HTTP and SSE streaming

Use this flow when all text is already available and audio should start before generation finishes.

Raw audio

with client.audio.speech.with_streaming_response.create(
    model="teen-v1",
    voice="default",
    input="Write audio while it is generated.",
    response_format="mp3",
) as response:
    response.stream_to_file("speech.mp3")

Use iter_bytes() instead of stream_to_file() when forwarding chunks to a player or downstream response.

SSE events

with client.audio.speech.with_streaming_response.create(
    model="teen-v1",
    voice="default",
    input="Return audio events.",
    response_format="mp3",
    stream_format="sse",
) as response:
    for event in response.iter_events():
        if event.type == "audio.delta":
            consume(event.audio)

SSE returns ordered audio.delta events followed by audio.done. Async clients provide async context managers and async for iteration.