生成字幕功能完全不断句，只会直接返回全文

[dearvcc]

🐛 Bug

根据文档，下面的命令可以生成字幕文件：
funasr tst.mp4 --output-format srt -o subs --lang zh

但结果是巨长的一整句，如图所示。

已经升到最新版：
funasr --version
funasr-1.3.17

[LauraGPT]

Thanks for the clear report. I reproduced the CLI-side cause: the top-level funasr ... --output-format srt path was not requesting sentence-level timestamps, and the default SenseVoice CLI config did not load a punctuation model for subtitle generation, so the SRT formatter could only fall back to one full-text cue.

I opened a fix in #3254. It makes SRT/TSV generation request sentence timestamps, loads ct-punc for subtitle output when needed, and keeps a valid timestamp/audio-bound fallback. The focused CLI/subtitle tests pass locally.

Before this PR is merged/released, a workaround is to use the subtitle example path that already requests sentence timestamps, for example:

python examples/subtitle/generate_subtitle.py tst.mp4 -o subs/tst.srt --lang zh