Issue with audio inference

[karlbernard2]

I'm trying to test your model, but I ran into an issue with the audio inference, maybe you would have some ideas what could be wrong?

`docker run --gpus all rlleshi/phar python src/demo/multimodial_demo.py /mnt/videos/tr_87505_hd.mp4 /mnt/videos/demo.mp4
Resizing video for faster inference...
Moviepy - Building video temp/tr_87505_hd.mp4.
MoviePy - Writing audio in tr_87505_hdTEMP_MPY_wvf_snd.mp3
MoviePy - Done.
Moviepy - Writing video temp/tr_87505_hd.mp4

Moviepy - Done !
Moviepy - video ready temp/tr_87505_hd.mp4
load checkpoint from local path: checkpoints/har/timeSformer.pth
Performing RGB inference...
100%|██████████| 12/12 [00:17<00:00, 1.44s/it]load checkpoint from local path: checkpoints/har/audio.pth
Performing audio inference...

8%|▊ | 1/12 [00:14<02:43, 14.88s/it]
Traceback (most recent call last):
File "src/demo/multimodial_demo.py", line 601, in
main()
File "src/demo/multimodial_demo.py", line 562, in main
audio_inference(clip, args.coefficients)
File "src/demo/multimodial_demo.py", line 385, in audio_inference
results = inference_recognizer(AUDIO_MODEL, out_feature)
File "/workspace/phar/mmaction2/mmaction/apis/inference.py", line 99, in inference_recognizer
raise RuntimeError('The type of argument video is not supported: '
RuntimeError: The type of argument video is not supported: <class 'str'>`

[rlleshi]

Hi karlbernard,
could you try to put a time.sleep(1) just above results = inference_recognizer(AUDIO_MODEL, out_feature), here?

subprocess.run is supposed to be blocking through.

[gauharbains]

I'm running into a similar issue. Does this ^ fix it ?

[gauharbains]

@rlleshi I tried this, but it didn't solve it.

[rlleshi]

@gauharbains could you please try to apply this fix in this PR here.

So just replace the glob search in mmaction2/tools/data/build_audio_features.py file with this line:

files = glob.glob(args.audio_home_path + '/*' * args.level + args.ext)

[gauharbains]

@rlleshi That worked for me, thanks!

[rlleshi]

Fixed at 4a14ea4