Synthesizing speech with bidirectional streaming

Amazon Polly provides a StartSpeechSynthesisStream operation that establishes an HTTP/2 connection with bidirectional communication between your application and the service. Text flows from your application to Amazon Polly while synthesized audio flows back. You send text as it becomes available, and Amazon Polly returns audio as it synthesizes, without either side waiting for the other to finish.

This is useful when text is produced progressively rather than all at once. For example, a customer service chatbot powered by a foundation model on Amazon Bedrock generates its response token by token. With bidirectional streaming, your application can forward each text chunk to Amazon Polly as the model produces it and begin playing audio back to the caller while the model is still generating the rest of the response.

This operation requires the generative engine and an AWS SDK that supports HTTP/2 event streams. The audio arrives as a sequence of chunks that your application accumulates into a complete audio output. Speech marks are not supported by this operation.