>We're aware it's likely that if you're reading out a long string of digits like this, you're likely to slip up or say a wrong digit. Mistakes are a very human feature, but as long as you correct it and carry on, they're fine - we just need evidence of humans having said the whole prime, not necessarily in one go. You can watch your video back through before submitting to check it's ok. If you want to edit together multiple clips to make your video, that's also fine!

So they need the some degree of semantic understanding to recognize corrections, in the various forms they can take. Speech to text alone is insufficient.