A half year ago I've already successfully trained multiple models using 'Audio + human labeled transcript' data in Microsoft speech portal. Now I wanted to train a new model but I only get the error message:
The base model isn't valid for this operation.
I tried to train the baseline model '20200115' and the older one '20190927'. Both failed with the error message above.
I already checked my audio data, which fits the requirements. The audio files are wav-files, 16-bit PCM, 1 channel.
I've also tried to create a new model using the old data which I had used to create the currently working model. This data is still available in speech portal so I had not to upload it again. For this I used the same baseline model '20190927' and got the error message too.
So Microsoft, is there anything you've changed? Is the website about data requirements still up to date? Or is this a bug?
What can i do to train a new model?
It's a known issue and a hotfix is currently being deployed. Because of the current situation the deployments takes longer than usual and will complete until the end of the week.