I experience issues with the TTS functionality to read out loud numbers. Specifically for medium and longer numbers (e.g. IBANs or similar), the model randomly removes numbers.
e.g. 123456789 results in 12378 or in another try 1356789.
A similar issue was raised that text to speech is dropping numbers when reading a numbered list, however I believe the defect is much bigger than only that.
Unfortunately, it is unacceptable behaviour because reading out what is provided is clear expectation to have proper quality.