Open MarkoMackic opened 2 months ago
Using speak_text
doesn't leave dangling threads ( stop_speaking_async
removed).
Using stop_speaking
instead of stop_speaking_async
causes segfault (start_speaking_async
used to initiate the speaking ). First thread exited successfully.
The problem is that _async
function require that the caller eventually calls result_future.get()
.
If this is by design, then at least it should be documented.
This item has been open without activity for 19 days. Provide a comment on status and remove "update needed" label.
This item has been open without activity for 19 days. Provide a comment on status and remove "update needed" label.
Issue is still present, and won't go away by it self. Very easy to reproduce.
Can we get any info about this from your side?
This item has been open without activity for 19 days. Provide a comment on status and remove "update needed" label.
Here is minimal reproduction example ( self contained so I don't have to follow all the instructions, you don't need anyio if you're using the pure_thread version, it's here just because we're using it, so I included it ). The problem here is that worker threads for MS speech synth ( the recognizer has the same problem ) are left dangling. The machines we're using are Ubuntu 22.04 ( also my machine ). We're using MS speech recog/synth in our system which handles many simultaneous real time audio connections, and we spawn a thread for each of those, and then inside it we run synth/recog . It seems by this data that 5 worker threads are spawned for each synth instance, so to sum it up, it quickly chokes the system.
If there is something I'm missing in how we use the API, please do let me know.
SDK version confirmed: 1.37.0, 1.38.0
Here is example of the output: