generated wav has part of ref wav

[sidoniepc]

Checks

• This template is only for bug reports, usage problems go with 'Help Wanted'.
• I have thoroughly reviewed the project documentation but couldn't find information to solve my problem.
• I have searched for existing issues, including closed ones, and couldn't find a solution.
• I am using English to submit this issue to facilitate community communication.

Environment Details

Hello, I was using cli inference, but I found when set speed=0.5, the output audio is mix of ref audio and audio of gen text

Steps to Reproduce

just run cli inference in document

✔️ Expected Behavior

Should only output only audio of gen text

❌ Actual Behavior

set speed=0.5, the output audio is mix of ref audio and audio of gen text

[SWivid]

You can check the model overview in https://swivid.github.io/F5-TTS/
The ref and gen parts are actually generated together, then ref part discarded.
When set speed 0.5 for long text, the ref gen estimated audio lengths severely mismatch.

[SWivid]

Feel free to open if further questions