shot sentences question,and dataset question.

[0x1001u]

Checks

• [X] This template is only for question, not feature requests or bug reports.
• [X] I have thoroughly reviewed the project documentation and read the related paper(s).
• [X] I have searched for existing issues, including closed ones, no similar questions.
• [X] I confirm that I am using English to submit this report in order to facilitate communication.

Question details

First question: Does a sampling rate of 16000 have any impact on the audio? Does it have to be 24000? The second issue is that when generating sentences of one or two words, it is not possible to generate them correctly. For example: Hello, thank you. May I ask how to adjust it?

[SWivid]

The current base model is trained on 24khz, so it needs an input with 24khz. For short sentence generation, the simply linear estimated duration is not accurate, thus consider fix duration (reference + to generate) , in seconds https://github.com/SWivid/F5-TTS/blob/e78ae2ce92ff1f2357af70e08b44fb09981ce1ce/src/f5_tts/infer/utils_infer.py#L51