When boto3 doesn't have it (yet), you write it: a real-time speech-to-speech story in Python
Saturday, May 30
12:25 - 12:55
In March 2026, AWS shipped bidirectional streaming for Amazon Polly—but not for Python’s boto3. This talk walks through a proof of concept that used the brand‑new HTTP/2 speech‑to‑speech API live, then hit the wall where the SDK stopped. We’ll unpack what could be reused, what had to be built from scratch, the messy detours that led to a new PyPI package, and the bigger question every Python team faces in this situation: do you publish your own library, or push changes upstream?
What happens when you need an AWS service that boto3 doesn’t expose yet?
In March 2026 AWS released bidirectional streaming for Amazon Polly: an HTTP/2 API for real-time speech-to-speech use cases. The Java SDK got it. boto3 didn’t.
After a brief intro to the goal of the PoC and how it works (likely running live in the room), we look at what got reused from other projects and what had to be written from scratch when boto3 came up short. From there, we dive into the Polly adventures: the detours, the dead ends, and the Python package on PyPI that came out the other side. And we’ll close with the question that sat behind this whole project, and that doesn’t answer itself: when you fill a gap like this, do you publish your own package on PyPI or push it upstream as a PR? In this case, both directions could be right, and we’ll lay them out side by side.
Alessandra Bilardi
Alessandra ha un’esperienza colorata: partita come sviluppatore, virato a operation e tornata a pieno regime sui dati a tutto tondo come al tempo accademico. Ora consulente in soluzioni di dati: dall’ingestion all’analisi e trasformazione, dalla predizione e generazione, alla visualizzazione per business e monitoring. Nel tempo libero organizza eventi locali per data scientist, developer e cloud engineer con AWS User Group Venezia, PyData Venice e PyVenice, ed è Coderdojo mentor in eventi locali per giovani coder.