How can I get acess to the TTS models?

I can confirm the Http response includes the headers Transfer-Encoding: chunked. I was able to stream the response in Node. But the initial latency is close to 1 second, so not ideal.