I'm using Azure's TTS service and was wondering if the rate and pitch values are applied to the audio files after generating a base sound, or are they part of the AI algorithm that generates the base sound?
I would like to know the above because it affects the choice of sampling (and available voices) for the generated files. For instance - if I generate a sound file at 24kHz, which is later stretched based on the settings (rate, pitch) there will be some loss of quality compared to a sound file generated at 48kHz which is then post processed. If, however, the rate, and pitch, are part of the AI algorithm the sound wont suffer quality artifacts.
This is also important because if the sound is processed at a later stage I could just do that with better accuracy in a DAW instead of relying on the settings in the TTS process. If however the algo modifies the sounds based on the settings in the TTS then the quality will be as good as it gets.