DeepSpeech's "Mandarin models trained on an internal corpus composed of 2000h of read speech"?

[botaurus]

(doing preliminary research so apologize for any ignorance)
DeepSpeech has a Mandarin model trained from 2000h but I don't think there is access to the source data (its described as "internal"). Is there an equivalent for coqui-ai? Common Voice has way less than 2000h of Mandarin data. Anyway to use a release model from Deepspeech in coqui-ai STT or covert it or something?
Thank you
https://github.com/mozilla/DeepSpeech/releases