Improving Arc Captions - Does it learn?

[lftapper]

Weird question incoming. Is there anyone knowledgable about the engine behind the Arc (Studio) captioning tool? While I'm fairly impressed already with Arc's captioning ability, I'm curious if it has the ability over time to get better at captioning. Does it learn from the videos we upload or the changes we make to the captions? Is it saving any of that data and improving its transcribing skill the more that we use it?

Thanks!

[JACOBSEN_C]

I was told, or read somewhere speachmatics is the service.  Sounds like a bit of AI is actually used and that it does learn...

"Speechmatics launches Automatic Linguist" Nov. 2017

[lftapper]

Thanks for getting me on the right track  @JACOBSEN_C ‌!

[James]

This may not be where you read it,  @JACOBSEN_C , but I found this news release about partnering with Speechmatics from 2017. It says that Canvas is using their speech-to-text transcription service.

The rest of this is educated speculation:

I think it would be necessary for Arc to send the transcripts back to the transcription service for any learning or improvement to happen. As it is, I think it's a separate, one-time, and one-way process. They transcribe the file and pass off a transcript to Arc, but Arc has its own caption editor and doesn't send changes back.

That is how most of the stand-alone services work. I looked into different services (Google, AWS, Watson) when I wasn't happy with Arc's translation editing capabilities and most of them worked by sending a file and getting back a transcript or captions.

[lftapper]

Thanks for the detailed reply  @James ‌! Unless I hear otherwise I'll take your educated guess as correct, since it confirms what I suspected. But it would be cool if it was two-way.