Inconsistent behavior on local and staging env when using RealtimeTTS with Azure Speech Services

[olivernaaris]

Hi

I'm using RealtimeTTS with socket.io to stream audio from Azure to backend (Python) and from backend to frontend (Next.js).
When using my code on my local dev env in MacBook - the audio streaming works well. However, when I deploy the backend code to a remote server (running inside Docker container) and frontend to Vercel - the audio synthesis just stops randomly without any error messages on the backend.
I also ran the Python backend in Docker on my local dev env to make sure that there is not MacBook vs Docker dependency issue.
It seems like RealtimeTTS audio streaming gets terminated without any error and it's impossible to tell why this is happening.
Any ideas?

I added the Socket.io and RealtimeTTS backend code with logs as attachments.
backend_logs.md
socketio_manager.md
text_to_speech_streaming.md

[olivernaaris]

Seems to me that the issue was related to Azure Speech SDK as I got the same issue when using just plain Azure SDK without RealtimeTTS. I found some issues in their GitHub that in theory are somewhat related, although I did not see exact correlation with my issue. For example: Azure-Samples/cognitive-services-speech-sdk#2280.
In the end I decided not to waste any more time on it and replaced Azure Speech SDK with AWS Polly. I got AWS Polly working in 1-2 days without issues.

[KoljaB]

Thanks for the update. I could not tell what the issue was, otherwise I would have answered. How is Polly, never tried it, are you satisfied with quality, latency etc?

[olivernaaris]

No problem. AWS Polly has less voices and accents then compared to Azure. However the voice quality is sufficient and latency is better than Azure. Overall it was less of a hassle to get AWS Polly to work as the documentation was pretty good and the SDK less buggy.