Hi. I have the same problem when trying to start audio classification with custom recipe. (Possible Bug in Web Application - #13 by Gaughar2)
That's how error looks like:
That's my recipe:
That's my command line:
python3 -m prodigy asr_level_test asr_task1 /home/prodigy/task_asr_level -F /home/prodigy/asr_level_recipe.py
That's my prodigy version:
Because we're using a
choice as our
view_id, we need to specify an options key which is currently missing in our loop:
for eg in stream:
eg["options"] = [
# some code...
For your case, you can probably (1) merge it into one list of
"options" and make it
"Action: [some option]", "Level: [some option]" or (2) use
choice_style=multiple in the config, and just do some validation on your end.
You can also check different ways to design your recipe in these threads:
Thank you for help!
I tried these ways and choose the one with multiple style. My task is to classify audios. Could you tell me, if there any way to save labels and meta in dataset but not to save labeled audios in dataset (that is in prodigy.db)? I don't want to copy there 80+ Gb audios
Hi @Adlinga !
There are two ways you can approach this:
- You can add some logic using the
before_db callback. You can check the example in this documentation and adapt it to your audio samples.
- Use the AudioServer loader.
Also, I highly-suggest to keep the original audio files because if they change, you'd lose the reference to the original data.
Yeah! Now it's perfect. Thank you again)