Support for Context Across Multiple Sentence Generation

Checks

• This template is only for research question, not usage problems, feature requests or bug reports.
• I have thoroughly reviewed the project documentation and read the related paper(s).
• I have searched for existing issues, including closed ones, no similar questions.
• I am using English to submit this issue to facilitate community communication.

Question details

Hi,

First of all, thank you for your great work on this model.

I have a question regarding inference with multiple text segments.

To improve generation speed, I split long texts into individual sentences and generate them separately. However, this causes the model to lose the overall context. As a result, each sentence is spoken as if it were independent, leading to unnatural prosody, abrupt transitions, and inconsistent intonation between sentences.

Is there currently a way to preserve or provide context when generating multiple segments separately? For example, something similar to:

• passing the previous sentence as context,
• maintaining an internal generation state,
• or any other mechanism that helps the model produce more coherent speech across sentence boundaries.

If this is not currently supported, are there any recommended workarounds or plans to add such functionality in the future?

Thank you!