Time length prediction method and device, speech synthesis method and device, model training method and device, medium and equipment

The embodiment of the invention discloses a duration prediction method and device, a speech synthesis method and device, a model training method and device, a medium and equipment. The method comprises the following steps: when time length prediction is carried out on text information of to-be-synthesized speech, phoneme features, word segmentation features and rhythm features are considered, so that each predicted time length is related to the phoneme features, the word segmentation features and the rhythm features; in this way, the prediction duration of the text information comprises the prediction duration of each phoneme and the prediction duration of each rhythm, so that the obtained prediction duration is more accurate, when speech synthesis is carried out, the synthesized speech comprises rhythm changes, the obtained synthesized speech is more accurate, and in addition, the speech synthesis efficiency is improved. The duration prediction model is separated from the acoustic model for independent train