Recordings intended for training the AI:
- should be PURE NOISE
- should NOT contain voice, cw, wsjt, sine wave tones, etc
- should be made with the radio's bandwidth filter set to 2.8KHz
- should be 30 minutes
- are not necessary if your noise is already handled perfectly (total silence at 100% filtered)
https://ournetplace.com/rm-noise/record ... -problems/
After you make a recording:
Here is what periodically happens:
- [change] You must inspect and SUBMIT your recordings to be included in the training dataset [/change]
- the noise files are manually examined for problems
- good recordings are added to the training dataset
- the training is performed and the latest AI filter is uploaded to the servers. The latest ssb voice filter will always be #1. The latest ssb cw filter will always be #2. No action is required at the client.
Find the status here:
I am able to perform the training about once per month. Each filter takes about a week. You can see the schedule on the system status page:
https://ournetplace.com/rm-noise/system-status/
My vision for the future includes a user portal where each user's recordings can be listed, visualized, played, and deleted if desired, however I can't even begin to offer a timeline.