So that between requests, for example, voicing a long lecture, the answers to the same prompt were given in the same voice.
Because now the prompt is the same - voicing the lecture text consists of 5-10 mp3 files (the same number of requests to the model) and every time I get a different result, it would be good to programmatically set one seed to each request so that the voice does not change.
It is this problem (the diversity of voices with each request) that makes - makes this API unusable