Kaldi Memory Leak

Hey everyone!

As stated in the title, i’m experiencing a memory leak with Kaldi. I’ve tried other Speech to Text modules, but they just don’t work as well.

As soon as i talk to my Raspberry, it takes 4-5 hours until the memory is overflowing and it freezes.
If i don’t say anything (i mostly need it for sheduled actions) it’s running fine for days.

HTOP is telling me rhasspyasr_kaldi_hermes is the one munching away, so i tried a few things to omit that. no success so far.

One potential cause for this is that i’ve made my own intent handling in python. In there i’ve made sure that the ASR stops after it’s duty is done twice:

	if topic.startswith('hermes/hotword/') and topic.endswith('/detected'):
		client.publish("hermes/asr/startListening", json.dumps({"stopOnSilence": "true"}))
		return

	if topic == "hermes/asr/textCaptured":
		client.publish("hermes/asr/stopListening", json.dumps({}))

But i suppose Kaldi keeping on listening isn’t the issue here.

I’ve also taken a look at MQTT-Explorer, but i haven’t recognized anything out of the ordinary over there
image_2023-04-17_215117038

Any ideas? Thank you for your help!