Clear Cache for file upload

Using Streamlit
21

Hey @hbredin,

Will check on this. Your still working with this app right?

Marisa

22

@Marisa_Smith Yes, this app. And am using the .wav audio files available here for testing: pyannote-audio/tests/data at develop · pyannote/pyannote-audio · GitHub

Those are 30seconds long so the visual output should also be 30s long


23

Hey @hbredin,

I have tested this out locally, and I am able to upload a whole audio file (33 sec in length) using Streamlit 0.76 and it plays fine when I pass it to st.audio. See my attached screen shot:

Screen Shot 2021-02-05 at 2.51.12 PM
Screen Shot 2021-02-05 at 2.51.12 PM1137×707 44.4 KB

@karriebear noticed that your passing the file object into Inference but we are unclear on what that does. What sort of processing are you doing on the audio?

vad = Inference(
    "hbredin/VoiceActivityDetection-PyanNet-DIHARD",
    batch_size=8,
    device="cpu",
)

Marisa

24

Thanks @Marisa_Smith @karriebear for looking into this!

(For some reason – I cannot post links in here. You’d have to add the github.com prefix yourself
)

Code of Inference class is here: pyannote/pyannote-audio/blob/977c3de8f7d062c7781b1b7ed45f38ba254b5118/pyannote/audio/core/inference.py#L44.

It starts by loading the whole file here: pyannote/pyannote-audio/blob/977c3de8f7d062c7781b1b7ed45f38ba254b5118/pyannote/audio/core/inference.py#L471

The above line actually calls torchaudio.load(file) here: pyannote/pyannote-audio/blob/977c3de8f7d062c7781b1b7ed45f38ba254b5118/pyannote/audio/core/io.py#L254

Could it possible that file_uploader provides a file handle to the rest of the code while it is still being uploaded? The behavior I expected that it would only return once the file is completely uploaded.

25

I have a similar issue.
I have two pages in my app and each of them accepts an image. Once I am done working on one page and move to the next, I still see the file that was uploaded on the previous page as the uploaded file on the current page.

I am not using any caching and using streamlit 0.76

1 Like
26

did you find a solution to this?