Custom Speech Commands Dataset

Help
#1

@ShawnHymel I tried your Custom Speech Commands Dataset
and Generate Features errored on 329 of the 530 files, aka they could not be ingested with an error like:

WARN: failed to process 129170/training/dracarys.audio(113).wav.3bdb4gqe.ingestion-6d97ddd674-8xjv5.json: Error

  • Flagged all archie (19 total files)
  • Flagged all fenrir (16 total files)
  • Flagged all goodnight (15 total files)
  • Flagged all hadouken (109 total files)
  • Flagged all hey_archie (15 total files)
  • Flagged all hey_fenrir (13 total files)
  • Flagged all how_are_you (22 total files)
  • Flagged 20 joke (46 total files)
  • Flagged 100 dracarys (100 total files)
  • Flagged none trick_or_treat (66 total files)

The Generate Features job finally failed.

Why are so many files being flagged as not ingestible?

Project ID: 29170

#2

Hi @MMarcial,

Where did you see this error? Was it on the Studio or were you running another script to upload?

I just tried the curation and upload process using this notebook, and it worked OK for me. https://github.com/ShawnHymel/ei-keyword-spotting/blob/master/ei-audio-dataset-curation.ipynb

Also, 29170 does not seem to be your project–is that the correct project ID on Edge Impulse?

#3

The injection error I saw 2 days ago was during Generate Features not during Upload.

I just ran Generate Features today and it worked as expected.

(I fat fingered the Project ID. It’s 129170 not 29170.)

1 Like
#4

Hmmm…that’s weird. Maybe it happened during an update (or better, a glitch in the Matrix :). Let me know if it happens again!