search

maxbbraun / whisper-edge

OpenAI Whisper for edge devices
MIT License
107 stars 16 forks source link

run.sh stops after "Warming model up..." #6

Open AnaCoda opened 11 months ago

AnaCoda commented 11 months ago

On my Jetson Nano Dev Kit 4GB, build.sh seemed to setup smoothly. However, when I run bash whisper-edge/run.sh, I get the following output:

[ .................... stream.py:73] Loading model "base.en"...
[ .................... stream.py:77] Warming model up....

There is no indication of any error but it stops there. There was also one time it only said [ .................... stream.py:73] Loading model "base.en"... Any idea why this may be happening? It seems to indicate an issue with whisper.transcribe(model=model, audio=np.zeros(block_size, dtype=np.float32))

Also, how long is "Warming model up" supposed to take? It seems to hang for half a minute before the program exits

horvathgergo commented 11 months ago

Did you test the microphone?

AnaCoda commented 11 months ago

Yes, the microphone was fine. It seems that the Jetson was running out of memory. A combination of closing Firefox/reducing time/adding swap memory seems to have fixed the problem.

However, this is not ideal and for some reason the inference seems to lag sometimes (some are very fast, some are twice as slow) even with the tiny.en model. Any idea why this may be happening?