Machine learning models

[Jeff AJ6LG]

Hi Mooneer,
I am curious about the training of the machine learning models used for FreeDV. Is the training ongoing, in the sense that a growing body of federated training data is used to update the models?  Are the new models then incorporated in each FreeDV application update?
-Jeff AJ6LG

[Mooneer Salem]

Hi Jeff,

As far as I know there hasn't been any additional training (other than what would be required as part of RADEV2 development). We've found that the 200 hours of training originally done for RADEV1 has been sufficient to cover most voices and band conditions. Hope this helps!

Thanks,

-Mooneer K6AQ

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.
To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/digitalvoice/65213fa8-a3a6-4ce2-af14-5b631b703c97n%40googlegroups.com.

[Jeff AJ6LG]

Hi Mooneer,

That makes sense.

Thanks!

-Jeff AJ6LG