Reading skips large chunks when there are multiple paragraphs

[Britnell]

Describe the bug so i entered a text of multiple paragraphs in the /web ui, and it will :

• read 1st sentence of 1st paragraph
• then skip to 2nd sentence of 2nd paragraph
• then skip to 3rd sentence of 3rd paragraph etc

using the api endpoint directly it :

• read the first paragraph correctlly entirely
• then skips to the last sentence of the 2nd paragraph
• then skips to the last sentence of the 3rd paragraph

Branch / Deployment used main branch, this is in docker-compose running setup

Operating System mac OS

Additional context hey team, thanks so much for this awesome project!! TTS is so awesome and its so nice to run it easily in docker container. Im combining with ReadAloud browser extension, as that has an option for openAi api which i replace w the localhost url

[Britnell]

/web ui ok i just found out that at higher speed it works correctly and reads out everything. anything above x1.3 works correct, however 1.0, 1.1, 1.2 all do the paragraph skip ...

[Britnell]

oh no it still skipped latter, but at some more erratic point ... weird

[RBEmerson970]

Under Docker, Win 11, i9&RTX4090, up to 10 minutes, 12 seconds (+/- 2 sec.) all material is read back completely. I've generated audio files running to approximately 3 hours, 45 minutes, again without dropouts or skipped chunks. I use FastKoko as a check on my fiction writing. While Kokoro has room for improvement, using FastKoko to get my work read back doesn't present significant errors such as dropouts.

[Britnell]

ah i switched to ./start-cpu.sh and it works seems i only get the issues with ./start-gpu_mac.sh